Kingyens Angela T 201211 PHD Thesis

BANKRUPTCY PREDICTION OF
COMPANIES IN THE RETAIL-APPAREL INDUSTRY

USING DATA ENVELOPMENT ANALYSIS

by

Angela Tsui-Yin Tran Kingyens

A thesis submitted in conformity with the requirements
for the degree of Doctor of Philosophy

Graduate Department of Chemical Engineering and Applied Chemistry
University of Toronto

Copyright by Angela Tsui-Yin Tran Kingyens 2012

ii

BANKRUPTCY PREDICTION OF COMPANIES IN THE RETAIL-APPAREL
INDUSTRY USING DATA ENVELOPMENT ANALYSIS

Angela Tsui-Yin Tran Kingyens
Degree of Doctor of Philosophy
Graduate Department of Chemical Engineering and Applied Chemistry
University of Toronto
2012

Abstract
Since 2008, the world has been in recession. As daily news outlets report, this crisis has
prompted many small businesses and large corporations to file for bankruptcy, which has grave
global social implications. Despite government intervention and incentives to stimulate the
economy that have put nations in hundreds of billions of dollars of debt, and have reduced the
prime rates to almost zero, efforts to combat the increase in unemployment rate as well as the
decrease in discretionary income have been troublesome. It is a vicious cycle: consumers are
apprehensive of spending due to the instability of their jobs and ensuing personal financial
problems; businesses are weary from the lack of revenue and are forced to tighten their
operations which likely translates to layoffs; and so on. Cautious movement of cash flows are
rooted in and influenced by the psychology of the players (stakeholders) of the game (society).
Understandably, the complexity of this economic fallout is the subject of much attention. And
while the markets have recovered much of the lost ground as of late, there is still great
opportunity to learn about all the possible factors of this recession, in anticipation of and bracing
for one more downturn before we emerge from this crisis. In fact, there is no time like today
more appropriate for research in bankruptcy prediction because of its relevance, and in an age
where documentation is highly encouraged and often mandated by law, the amount and
accessibility of data is paramount an academics paradise!
iii

The main objective of this thesis was to develop a model supported by Data Envelopment
Analysis (DEA) to predict the likelihood of failure of US companies in the retail-apparel industry
based on information available from annual reports specifically from financial statements and
their corresponding Notes, Managements Discussion and Analysis, and Auditors Report. It
was hypothesized that the inclusion of variables which reflect managerial decision-making and
economic factors would enhance the predictive power of current mathematical models that
consider financial data exclusively. With a unique and comprehensive dataset of 85 companies,
new metrics based on different aspects of the annual reports were created then combined with a
slacks-based measure of efficiency DEA model and modified layering classification technique to
capture the multidimensional complexity of bankruptcy. This approach proved to be an effective
prediction tool, separating companies with a high risk of bankruptcy from those that were
healthy, with a reliable accuracy of 80% an improvement over the widely-used Altman
bankruptcy model having 70%, 58% and 50% accuracy when predicting cases today, from one
year back and from two years back, respectively. It also provides a probability of bankruptcy
based on a second order polynomial function in addition to targets for improvement, and was
designed to be easily adapted for analysis of other industries. Finally, the contributions of this
thesis benefit creditors with better risk assessment, owners with time to improve current
operations as to avoid failure altogether, as well as investors with information on which healthy
companies to invest in and which unhealthy companies to short.

Acknowledgements
The greatest compliment I have ever received is that I am a reflection of my
friends and mentors. This PhD has been an incredible journey
for because I had the privilege of learning from the wisest, working with the most brilliant,
sharing with the most generous, and laughing with the funniest individuals along the way.
First, I wish to express my deepest appreciation to
Paradi, for his support and guidance, and more importantly, for his offerings of ever
lessons on life. Being his student has been a gift:
understanding of my need to follow my
great things.
I would like to thank Professors Yuri Lawryshyn and Bradley Saville for always finding
the time to offer valuable feedback
Podinovski for serving on my defense committee
Department of Chemical Engineering and Applied Chemistry, the Faculty of Applied Science
and Engineering, and the Leaders of Tomorrow Office for their continued support in my personal
and professional development
and the Centre for Management of Technology and Entrepreneurship
I wish to extend a personal thanks to my family. In particular, I am indebt
incredible parents: my father, my hero, who inspires me through his hard work and
accomplishments, and my mother, my rock, who encourages me through her beauty and
compassion. Thank you both for your unconditional love
so lucky. Also, I would like to thank my brother, Chris, for j
without question.
And last, but never least, I thank my wonderful hu
me, for all that he is to me, and all that he will be. I am truly blessed.

iv
The greatest compliment I have ever received is that I am a reflection of my
s PhD has been an incredible journey one that I am extremely grateful
I wish to express my deepest appreciation to my advisor,
Paradi, for his support and guidance, and more importantly, for his offerings of ever
Being his student has been a gift: I am extremely grateful for his kindness, his
need to follow my heart to California, and his belief in my ability to do
thank Professors Yuri Lawryshyn and Bradley Saville for always finding
the time to offer valuable feedback, as well as Professors Mark Kortschot, Roy Kwon and Victor
Podinovski for serving on my defense committee. I would also like to acknowledge the
he Leaders of Tomorrow Office for their continued support in my personal
and professional development. I would also like to thank my friends at the
entre for Management of Technology and Entrepreneurship, present and alumni.
I wish to extend a personal thanks to my family. In particular, I am indebt
passion. Thank you both for your unconditional love, and putting my happiness first.
lucky. Also, I would like to thank my brother, Chris, for just always being there for me
And last, but never least, I thank my wonderful husband Jeff, for all that he has been for
me, for all that he is to me, and all that he will be. I am truly blessed.
The greatest compliment I have ever received is that I am a reflection of my wonderful family,
one that I am extremely grateful
my advisor, Professor Joseph C.
Paradi, for his support and guidance, and more importantly, for his offerings of ever-colourful
ul for his kindness, his
, and his belief in my ability to do
thank Professors Yuri Lawryshyn and Bradley Saville for always finding
as well as Professors Mark Kortschot, Roy Kwon and Victor
like to acknowledge the
he Leaders of Tomorrow Office for their continued support in my personal
like to thank my friends at the University of Toronto
, present and alumni.
I wish to extend a personal thanks to my family. In particular, I am indebted to my
my happiness first. I am
ust always being there for me
sband Jeff, for all that he has been for

Angela Tran Kingyens
May 2012
v

Table of Contents
ABSTRACT II
ACKNOWLEDGEMENTS IV
TABLE OF CONTENTS V
LIST OF TABLES X
LIST OF FIGURES XIII
LIST OF ABBREVIATIONS XVIII
LIST OF SYMBOLS XIX
1 INTRODUCTION 1
2 BACKGROUND I: ACCOUNTING AND FINANCIAL STATEMENT ANALYSIS 3
2.1 THE ANNUAL REPORT 3
2.1.1 BALANCE SHEET 3
2.1.2 INCOME STATEMENT 3
2.1.3 CASH FLOW STATEMENT 4
2.1.4 NOTES TO THE FINANCIAL STATEMENTS 6
2.1.5 MANAGEMENTS DISCUSSION AND ANALYSIS (MD&A) 8
2.1.6 AUDITORS REPORT 8
2.2 FINANCIAL STATEMENT ANALYSIS: RATIOS 9
2.2.1 INDUSTRY ANALYSIS 9
2.2.2 PROFITABILITY ANALYSIS 10
2.2.3 SHORT-TERM LIQUIDITY AND LONG-TERM SOLVENCY 11
2.3 BANKRUPTCY 12
2.3.1 ALTMANS BANKRUPTCY MODEL 13
2.3.2 OHLSONS BANKRUPTCY MODEL 15
2.4 FRAUD 17
2.4.1 INVESTIGATING FRAUD WITH FINANCIAL MEASURES 17
2.4.2 INVESTIGATING FRAUD WITH NON-FINANCIAL DATA 20
2.4.3 CRITICAL RED FLAGS 21
2.4.4 SUMMARY OF FRAUD LITERATURE 24
2.5 MARKET BETA EQUITY RISK AND OTHER FACTORS 25
vi

3 BACKGROUND II: PROFILE OF THE RETAIL-APPAREL INDUSTRY 28
3.1 OVERVIEW 28
3.2 SWOT ANALYSIS 30
3.2.1 STRENGTHS 30
3.2.2 OPPORTUNITIES 31
3.2.3 WEAKNESSES AND THREATS 32
3.3 FUTURE OUTLOOK 35
4 DATA ENVELOPMENT ANALYSIS 36
4.1 INTRODUCTION 36
4.2 CCR MODEL 38
4.2.1 INPUT ORIENTATION 38
4.2.2 OUTPUT ORIENTATION 41
4.3 BCC MODEL 42
4.3.1 INPUT ORIENTATION 43
4.3.2 OUTPUT ORIENTATION 44
4.4 ADDITIVE MODEL 45
4.5 SLACKS-BASED MEASURE OF EFFICIENCY MODEL 46
4.6 SENSITIVITY ANALYSIS 48
4.7 SUMMARY OF THE BASIC DEA MODELS 48
4.8 DEA CHARACTERISTICS 48
4.8.1 STRENGTHS 48
4.8.2 LIMITATIONS 50
4.9 DEA STUDIES IN BANKRUPTCY 52
5 OBJECTIVES 55
6 DATA COLLECTION, PREPARATION AND EXPLORATION 56
6.1 LIST OF RETAIL-APPAREL COMPANIES 56
6.2 DEFINITION OF STUDY PERIOD, COMPARABLE UNITS AND COMPANY STATE 57
6.3 FINANCIAL DATA 60
6.3.1 LIST OF FINANCIAL VARIABLES 60
6.3.2 COLLECTION AND ORGANIZATION OF FINANCIAL VARIABLES 60
6.3.3 DISTRIBUTIONS OF FINANCIAL VARIABLES 64
6.3.4 CORRELATIONS OF FINANCIAL VARIABLES 67
6.3.5 PROFITABILITY ANALYSIS 69
6.3.6 SHORT-TERM LIQUIDITY AND LONG-TERM SOLVENCY 72
6.3.7 EARNINGS MANIPULATION RISK 73
6.3.8 QUALITY OF INCOME 74
6.4 MANAGERIAL DECISION-MAKING DATA 74
vii

6.5 OUTLIERS AND MISSING DATA 77
6.6 ECONOMIC DATA 78
6.7 MARKET AND STOCK PERFORMANCE 80
6.8 CROSS-DATA EXPLORATION 83
6.8.1 CORRELATION BETWEEN FINANCIAL AND MANAGERIAL DECISION-MAKING DATA 83
6.8.2 CORRELATION BETWEEN FINANCIAL AND ECONOMIC DATA 85
6.8.3 CORRELATION BETWEEN MANAGERIAL DECISION-MAKING AND ECONOMIC DATA 86
6.8.4 CORRELATION TO MARKET AND STOCK PERFORMANCE DATA 87
6.9 SUMMARY OF DATA EXPLORATION 89
7 BENCHMARKS 91
7.1 DEFINITION OF TYPE I AND TYPE II ERRORS 91
7.2 ALTMANS Z-SCORE 92
7.3 CREDIT RATING AGENCIES 93
8 METHODOLOGY 94
8.1 APPLICATION OF DATA ENVELOPMENT ANALYSIS TO BANKRUPTCY PREDICTION 94
8.1.1 FRONTIERS, INPUTS AND OUTPUTS 94
8.1.2 A NEW CLASSIFICATION APPROACH: A MODIFIED LAYERING TECHNIQUE 95
8.2 OVERVIEW: METRICS AND THE SECOND STAGE MODEL 96
8.3 VARIABLE SELECTION AND METRICS TESTED 97
8.4 SLACKS-BASED MEASURE OF EFFICIENCY (SBM) MODEL 99
8.5 CLASSIFICATION BY ZONES AND LAYERING 100
9 PRELIMINARY MODELS: RESULTS AND DISCUSSION 102
9.1 PRINCIPAL COMPONENT ANALYSIS AND LOGISTIC REGRESSION METRICS 102
9.2 PROFITABILITY METRICS 106
9.3 COMBINING NORMAL AND INVERSE SCORES 110
9.4 INSOLVENCY METRICS 113
9.5 SUMMARY OF PRELIMINARY MODELS 115
10 FINANCIAL STATEMENT AND MANAGERIAL DECISION-MAKING MODELS:
RESULTS AND DISCUSSION 116
10.1 INCOME STATEMENT METRIC 116
10.2 BALANCE SHEET METRICS 117
10.3 CASH FLOW STATEMENT METRIC 120
10.4 MANAGERIAL DECISION-MAKING METRIC 120
10.5 TRADE DATA METRIC 121
viii

10.6 CORRELATION BETWEEN METRICS 122
10.7 THE EFFECT OF INFLATION 122
11 MARKET AND ECONOMIC FACTORS: RESULTS AND DISCUSSION 125
11.1 GENERAL ECONOMIC INDICATOR 125
11.2 APPAREL INDUSTRY INDICATOR 126
11.3 HOUSING MARKET INDICATOR 127
11.4 PRICES INDICATOR 128
11.5 GENERAL MARKET PERFORMANCE INDICATOR 129
11.6 ALL-ENCOMPASSING MARKET AND ECONOMIC (ME) INDICATOR 129
11.7 COMBINING MARKET AND ECONOMIC FACTORS WITH FINANCIAL AND MANAGERIAL
DECISION-MAKING DATA 130
12 SECOND STAGE DEA MODELS: RESULTS AND DISCUSSION 132
12.1 MOTIVATION 132
12.2 LAYERED SCORES AND SECOND STAGE MODEL SPECIFICATIONS 134
12.3 CLASSIFICATION BY ZONES 135
12.4 CLASSIFICATION BY LAYERING 136
12.4.1 SECOND STAGE MODEL WITH IS, BSA, BSL AND MDM 137
12.4.2 SECOND STAGE MODEL WITH IS, BSA, BSL AND MDM WITH CFO 140
12.5 TYPE I AND TYPE II ERROR TRADE-OFF 140
12.6 COMPARING THE LAYERING AND NON-LAYERING APPROACHES 142
12.7 COMPARISON TO ALL-ENCOMPASSING MARKET AND ECONOMIC INDICATOR 143
12.8 IMPROVING PREDICTION BY CHANGING SAMPLE SIZE 145
12.9 TRANSLATING DEA RESULTS INTO TARGETS OF IMPROVEMENT 146
13 PREDICTION AND VALIDATION OF THE SECOND STAGE DEA MODEL 158
13.1 ESTIMATING THE PROBABILITY OF BANKRUPTCY 158
13.2 RELIABILITY AND VALIDITY OF THE SECOND STAGE MODEL 160
13.2.1 PREDICTION OF HEALTH AFTER 2009 BASED ON DATA FROM 1996 TO 2009 162
13.2.2 PREDICTION OF HEALTH AFTER 2008 BASED ON 1996-2008 DATA 162
14 CONCLUSIONS 165
14.1 SUMMARY OF MAJOR CONTRIBUTIONS 165
ix

14.2 RECOMMENDATIONS FOR FUTURE WORK 167
REFERENCES 169
APPENDIX A PARAMETRIC FRONTIER METHODOLOGIES 175
A.1 STOCHASTIC FRONTIER ANALYSIS (SFA) 175
A.2 THICK FRONTIER APPROACH (TFA) 175
A.3 DISTRIBUTION-FREE APPROACH (DFA) 176
APPENDIX B EXAMPLES OF DATA ENVELOPMENT ANALYSIS 177
B.1 ONE INPUT, ONE OUTPUT 177
B.2 MINIMIZE TWO INPUTS, UNITIZED OUTPUT 178
B.3 MAXIMIZE TWO OUTPUTS, UNITIZED INPUT 180
APPENDIX C REASONS FOR BANKRUPTCY AS CITED BY PRESS RELEASES 182
APPENDIX D SUPPLEMENTARY BACKGROUND 185
D.1 CREDIT RATING AGENCIES 185
APPENDIX E LIST OF METRICS 186
APPENDIX F MODELLING RESULTS 189
F.1 LOGISTIC REGRESSION (LR) METRIC 189
F.2 INCOME STATEMENT (IS) METRIC 190
F.3 BALANCE SHEET ASSETS (BSA) METRIC 191
F.4 BALANCE SHEET LIABILITIES (BSL) METRIC 192
F.5 MANAGERIAL DECISION-MAKING (MDM) METRIC 193
F.6 SECOND STAGE MODELS 194
F.7 PROJECTIONS FOR ACTIVE COMPANIES 196
F.8 SECOND STAGE MODEL RESULTS WITH DIFFERENT TIME WINDOWS 202
F.9 PROBABILITY FUNCTIONS FOR DIFFERENT TIME WINDOWS 204

x

List of Tables
TABLE 2.1: COMMON ENTRIES IN A BALANCE SHEET (GILL, 1994) 3
TABLE 2.2: COMMON ENTRIES IN AN INCOME STATEMENT (GILL, 1994) 4
TABLE 2.3: FINANCIAL ACTIVITIES REPORTED IN A CASH FLOW STATEMENT 5
TABLE 2.4: INDUSTRY ANALYSIS (STICKNEY ET AL., 2007) 10
TABLE 2.5: COMMON LIQUIDITY AND SOLVENCY RATIOS 12
TABLE 2.6: Z-SCORES BY FIRM TYPE AND ZONES OF DISCRIMINATION (ALTMAN, 1968; HANSON,
2003) 14
TABLE 2.7: ERROR CLASSIFICATION FOR ALTMANS ORIGINAL MODEL 14
TABLE 2.8: ERROR CLASSIFICATION FOR OHLSONS MODEL 16
TABLE 2.9: SAS NO. 82 (APOSTOLOU, 2001) 20
TABLE 2.10: DEFINITION OF OF SECURITY OR COMPANY 26
TABLE 2.11: MARKET RATIOS 27
TABLE 3.1: RETAIL INDUSTRY SEGMENTATION (US CENSUS, 2002) 28
TABLE 3.2: VALUE-ADDED BY INDUSTRY, 2005-2010, IN BILLIONS OF DOLLARS (BEA) 29
TABLE 3.3: MANUFACTURING AND TRADE (BEA) 29
TABLE 3.4: TOTAL SALES OF APPAREL INDUSTRY FROM 2001-2009 (US DEPARTMENT OF
COMMERCE) 29
TABLE 3.5: US POPULATION PROJECTIONS (US DEPARTMENT OF COMMERCE) 32
TABLE 3.6: BANKRUPTCY CASES COMMENCED, TERMINATED AND PENDING DURING TWELVE
MONTH PERIODS ENDING DECEMBER 31 35
TABLE 4.1: SUMMARY OF MODEL CHARACTERISTICS (COOPER ET AL., 2007) 48
TABLE 6.1: LIST OF COMPANIES BY TICKER CODE IN RETAIL-APPAREL INDUSTRY 56
TABLE 6.2: REASONS FOR CHAPTER 11 FILINGS CITED IN PRESS RELEASES 57
TABLE 6.3: POSSIBLE DMU STATES 59
TABLE 6.4: FINANCIAL ITEMS 60
TABLE 6.5: HIGHLY CORRELATED FINANCIAL ACCOUNTS ON THE BALANCE SHEET 68
TABLE 6.6: HIGHLY CORRELATED FINANCIAL ACCOUNTS ON THE INCOME STATEMENT 68
TABLE 6.7: HIGHLY CORRELATED FINANCIAL ACCOUNTS ACROSS STATEMENTS 69
TABLE 6.8: AVERAGE MEDIAN ROAS, PMS, TATS BY STATE 70
TABLE 6.9: AVERAGES OF RATIOS IN PROFIT MARGIN BREAKDOWN 71
TABLE 6.10: AVERAGES OF RATIOS IN ASSETS TURNOVER (TO) BREAKDOWN 71
TABLE 6.11: AVERAGE OF MEDIAN ROCE 71
TABLE 6.12: COMPANY LEVERAGE POSITION 72
TABLE 6.13: AVERAGE LIQUIDITY RATIOS 72
TABLE 6.14: AVERAGE LIQUIDITY RATIOS 72
TABLE 6.15: AVERAGE SOLVENCY RATIOS 72
TABLE 6.16: AVERAGE BENEISH INDEX 74
TABLE 6.17: BANKRUPT COMPANIES AND EARNINGS MANIPULATION RISK 74
TABLE 6.18: QUALITY OF INCOME 74
TABLE 6.19: MANAGERIAL DECISION-MAKING VARIABLES 76
TABLE 6.20: CHI-SQUARE TEST RESULTS OF MANAGERIAL DECISION-MAKING DATA AND STATE
77
TABLE 6.21: LIKELIHOOD OF MANAGEMENT TURNOVER 77
TABLE 6.22: PROBABILITY OF AUDITOR CHANGE BY STATE 77
TABLE 6.23: PERCENTAGE OF DMUS WITH NON-SERIOUS AND SERIOUS LEGAL PROCEEDINGS BY
STATE 77
TABLE 6.24: ECONOMIC FACTORS IN THIS THESIS 79
xi

TABLE 6.25: SPEARMAN CORRELATION COEFFICIENTS OF ECONOMIC FACTORS 80
TABLE 6.26: CORRELATION BETWEEN COMPOSITE INDICES 81
TABLE 6.27: AVERAGE ANNUAL RETURN, VOLATILITY AND BETA 82
TABLE 6.28: CORRELATION BETWEEN COMPOSITE INDICES AND STOCK PERFORMANCE 83
TABLE 6.29: CORRELATION BETWEEN FINANCIAL AND ECONOMIC VARIABLES 86
TABLE 6.30: CORRELATION BETWEEN MANAGERIAL-DECISION MAKING AND ECONOMIC
VARIABLES 87
TABLE 6.31: CORRELATION BETWEEN MARKET AND STOCK PERFORMANCE AND FINANCIAL
DATA 88
TABLE 6.32: CORRELATION BETWEEN MARKET AND STOCK PERFORMANCE AND ECONOMIC
FACTORS 89
TABLE 6.33: CORRELATION BETWEEN MARKET AND STOCK PERFORMANCE AND MANAGERIAL
DECISION-MAKING DATA 89
TABLE 7.1: CONFUSION MATRIX 91
TABLE 7.2: AVERAGE BANKRUPTCY SCORES OF RETAIL-APPAREL COMPANIES BY STATE 92
TABLE 7.3: TYPE II ERROR 93
TABLE 7.4: CUMULATIVE HISTORIC CORPORATE DEFAULT RATES (U.S. MUNICIPAL BOND
FAIRNESS ACT, 2008) 93
TABLE 8.1: COMMON VARIABLES MEASURING GOOD PERFORMANCE OR SUCCESS 98
TABLE 8.2: COMMON VARIABLES MEASURING POOR PERFORMANCE OR FAILURE 98
TABLE 8.3: ABBREVIATED LIST OF METRICS TESTED 99
TABLE 9.1: NORMAL SCORES CALCULATED WITH VARIABLES DERIVED FROM PCA BY STATE 102
TABLE 9.2: CONFUSION MATRIX FOR PCA NORMAL MODEL 103
TABLE 9.3: INVERSE SCORES CALCULATED WITH VARIABLES DERIVED FROM PCA BY STATE 104
TABLE 9.4: CONFUSION MATRIX FOR PCA INVERSE MODEL 104
TABLE 9.5: SUMMARY OF RESULTS FOR MODELS WITH VARIABLES SELECTED BY PCA AND LR 106
TABLE 9.6: CORRELATION BETWEEN PCA AND LR SCORES 106
TABLE 9.7: CORRELATION BETWEEN NORMAL AND INVERSE PROFITABILITY MODELS 108
TABLE 9.8: EXAMPLE OF INCONSISTENT NORMAL AND INVERSE MODELS OF A TWO INPUTS
TWO OUTPUTS SYSTEM 110
TABLE 9.9: PARALLEL CLASSIFICATION OF MODELS A AND B 111
TABLE 9.10: SERIAL CLASSIFICATION OF MODELS A AND B (NORMAL THEN INVERSE DEA) 112
TABLE 9.11: SERIAL CLASSIFICATION OF MODELS A AND B (INVERSE THE NORMAL DEA) 113
TABLE 9.12: PARALLEL CLASSIFICATION OF NORMAL AND WORST-PRACTICE MODELS 114
TABLE 9.13: SERIAL CLASSIFICATION OF NORMAL THEN WORST-PRACTICE MODELS 114
TABLE 9.14: SERIAL CLASSIFICATION OF WORST-PRACTICE THEN NORMAL MODELS 114
TABLE 10.1: CORRELATION BETWEEN METRICS AFTER FIRST RUN 122
TABLE 11.1: CORRELATIONS TO AGGREGATE GENERAL ECONOMIC INDICATOR 126
TABLE 11.2: CORRELATIONS TO AGGREGATE APPAREL INDUSTRY INDICATOR 127
TABLE 11.3: CORRELATIONS TO AGGREGATE HOUSING PERFORMANCE INDICATOR 128
TABLE 11.4: CORRELATIONS TO AGGREGATE PRICES INDICATOR 129
TABLE 12.1: INPUT-ORIENTED MODELS: SPECIFICATIONS AND RESULTS WITH INPUT VALUE
RANGE BETWEEN 1 AND 3 133
TABLE 12.2: INPUT-ORIENTED MODELS: SPECIFICATIONS AND RESULTS WITH INPUT VALUE
RANGE BETWEEN 1 AND 100 133
TABLE 12.3: CORRELATION OF LAYERED SCORES 134
TABLE 12.4: SECOND STAGE MODEL PREDICTION WITH CLASSIFICATION BY ZONES 135
TABLE 12.5: ACCURACY OF CLASSIFICATION BY LAYERING OF SECOND STAGE MODEL AND
INDIVIDUAL METRICS 139
xii

TABLE 12.6: SECOND STAGE MODEL WITH AND WITHOUT CFO METRIC 140
TABLE 12.7: SPECIFICITY AND SENSITIVITY TRADE-OFF 142
TABLE 12.8: PERFORMANCE OF NON-LAYERING AND LAYERING TECHNIQUES 143
TABLE 12.9: TESTS OF SIMILARITY BETWEEN SECOND STAGE SCORES AND ME INDICATOR 144
TABLE 12.10: DIRECTION OF CHANGE FROM YEAR TO YEAR 145
TABLE 13.1: PROBABILITIES OF BANKRUPTCY (B) AND NON-BANKRUPTCY (NB) 159
TABLE 13.2: ACCURACY OF CLASSIFICATION BY LAYERING OF SECOND STAGE MODEL FROM
ONE YEAR BACK 161
TABLE 13.3: SECOND ORDER POLYNOMIAL FUNCTIONS ESTIMATING THE PROBABILITY OF
BANKRUPTCY 161
TABLE B.1: TWO INPUTS AND ONE OUTPUT CASE 178
TABLE B.2: ONE INPUT AND TWO OUTPUTS CASE 180
TABLE D.1: BROAD CATEGORIES FOR ECONOMIC INDICATORS 185
TABLE D.2: RATING CODES FOR EACH CREDIT RATING AGENCIES 185
TABLE E.1: VARIABLES FROM PRINCIPAL COMPONENT ANALYSIS AND LOGISTIC REGRESSION 186
TABLE E.2: PROFITABILITY AND SUCCESS METRICS (NORMAL DEA FRONTIER) 186
TABLE E.3: PROFITABILITY AND SUCCESS METRICS (INVERSE DEA FRONTIER) 186
TABLE E.4: BANKRUPTCY AND FAILURE METRICS (WORST-PRACTICE DEA FRONTIER) 187
TABLE E.5: BANKRUPTCY AND FAILURE METRICS (BAD OUTPUTS MODEL) 187
TABLE E.6: BALANCE SHEET METRICS (NORMAL DEA FRONTIER) 187
TABLE E.7: INCOME STATEMENT METRICS (NORMAL DEA FRONTIER) 187
TABLE E.8: QUALITATIVE METRICS (NORMAL DEA FRONTIER) 187
TABLE E.9: VARIABLES FOR MEGA MODEL 188
TABLE F.1: NORMAL SCORES CALCULATED WITH VARIABLES DERIVED FROM LR BY STATE 189
TABLE F.2: CONFUSION MATRIX FOR LR NORMAL MODEL 189
TABLE F.3: INVERSE SCORES CALCULATED WITH VARIABLES DERIVED FROM LR BY STATE 189
TABLE F.4: CONFUSION MATRIX FOR LR INVERSE MODEL 190

xiii

List of Figures
FIGURE 2.1: RELATIONSHIP BETWEEN THE BALANCE SHEET AND CASH FLOWS 4
FIGURE 2.2: CAPITAL ASSET PRICING MODEL (CAPM) 26
FIGURE 2.3: CAPM SECURITY MARKET LINE 26
FIGURE 3.1: US IMPORTS AND EXPORTS BY YEAR (IN BILLIONS OF DOLLARS) (US DEPARTMENT
OF COMMERCE) 30
FIGURE 3.2: CONSUMER CONFIDENCE INDEX BY YEAR (1985 = 100) (THE CONFERENCE BOARD) 33
FIGURE 3.3: APPARELS SHARE OF TOTAL PERSONAL CONSUMPTION EXPENDITURES BY YEAR (IN
%) (US DEPARTMENT OF COMMERCE) 33
FIGURE 3.4: CONSUMER PRICE INDICES (1982-84 = 100) (BUREAU OF LABOUR STATISTICS) 34
FIGURE 3.5: US APPAREL INDUSTRY EMPLOYMENT BY YEAR (PRODUCTION WORKERS, IN
MILLIONS) (BUREAU OF LABOUR STATISTICS) 34
FIGURE 4.1: DATA ENVELOPMENT ANALYSIS VERSUS REGRESSION ANALYSIS 37
FIGURE 4.2: INEFFICIENCIES AND PROJECTIONS IN CCR AND BCC MODELS WITH ONE INPUT AND
ONE OUTPUT 38
FIGURE 4.3: DEA SYSTEM WITH MULTIPLE INPUTS AND MULTIPLE OUTPUTS 38
FIGURE 4.4: INPUT MINIMIZATION, OUTPUT MAXIMIZATION AND SLACKS 40
FIGURE 4.5: CONSTANT VERSUS VARIABLE RETURNS-TO-SCALE (RTS) 43
FIGURE 4.6: ADDITIVE MODEL 45
FIGURE 4.7: TRANSLATION INVARIANCE IN THE ADDITIVE MODEL 46
FIGURE 4.8: TRUE (THEORETICAL) FRONTIER VERSUS DEA (EMPIRICAL) FRONTIER 49
FIGURE 4.9: ONE INPUT, TWO OUTPUT EXAMPLE FOR USE OF RATIOS IN DEA 52
FIGURE 5.1: CURRENT LIMITATIONS 55
FIGURE 6.1: TIME FRAME OF AVAILABLE DATA FOR COMPANIES AENA-JWN 58
FIGURE 6.2: TIME FRAME OF AVAILABLE DATA FOR COMPANIES KDRH-ZUMZ 59
FIGURE 6.3: CREATION OF TEMPLATE FINANCIAL STATEMENTS 61
FIGURE 6.4: EXAMPLE OF TEMPLATE INCOME STATEMENT (IN $M) 62
FIGURE 6.5: EXAMPLE OF TEMPLATE BALANCE SHEET (IN $M) 62
FIGURE 6.6: EXAMPLE OF TEMPLATE CASH FLOW STATEMENT (IN $M) 63
FIGURE 6.7: COMBINING ALL DATA INTO ONE WORKSHEET (IN $M) 64
FIGURE 6.8: HISTOGRAM FOR AVERAGE ANNUAL CASH 64
FIGURE 6.9: HISTOGRAM FOR AVERAGE ANNUAL AR 64
FIGURE 6.10: HISTOGRAM FOR AVERAGE ANNUAL INV 65
FIGURE 6.11: HISTOGRAM FOR AVERAGE ANNUAL CA 65
FIGURE 6.12: HISTOGRAM FOR AVERAGE ANNUAL PPE 65
FIGURE 6.13: HISTOGRAM FOR AVERAGE ANNUAL AD 65
FIGURE 6.14: HISTOGRAM FOR AVERAGE ANNUAL TA 65
FIGURE 6.15: HISTOGRAM FOR AVERAGE ANNUAL AP 65
FIGURE 6.16: HISTOGRAM FOR AVERAGE ANNUAL CM 65
FIGURE 6.17: HISTOGRAM FOR AVERAGE ANNUAL CL 65
FIGURE 6.18: HISTOGRAM FOR AVERAGE ANNUAL LTD 66
FIGURE 6.19: HISTOGRAM FOR AVERAGE ANNUAL TL 66
FIGURE 6.20: HISTOGRAM FOR AVERAGE ANNUAL RE 66
FIGURE 6.21: HISTOGRAM FOR AVERAGE ANNUAL SE 66
FIGURE 6.22: HISTOGRAM FOR AVERAGE ANNUAL REV 66
FIGURE 6.23: HISTOGRAM FOR AVERAGE ANNUAL COGS 66
FIGURE 6.24: HISTOGRAM FOR AVERAGE ANNUAL SGA 66
FIGURE 6.25: HISTOGRAM FOR AVERAGE ANNUAL II 66
xiv

FIGURE 6.26: HISTOGRAM FOR AVERAGE ANNUAL IE 67
FIGURE 6.27: HISTOGRAM FOR AVERAGE ANNUAL NI 67
FIGURE 6.28: HISTOGRAM FOR AVERAGE ANNUAL CFO 67
FIGURE 6.29: HISTOGRAM FOR AVERAGE ANNUAL CFF 67
FIGURE 6.30: HISTOGRAM FOR AVERAGE CFF 67
FIGURE 6.31: HISTOGRAM FOR AVERAGE ANNUAL CC 67
FIGURE 6.32: MEDIAN EFFICIENCY AND PROFITABILITY CHART FOR ALL COMPANIES 70
FIGURE 6.33: STRATEGY FOR DERIVING A THRESHOLD FOR EVALUATING RATIOS 73
FIGURE 6.34: EXAMPLE OF ORGANIZATION OF QUALITATIVE INFORMATION EXTRACTED FROM
ANNUAL REPORTS 76
FIGURE 6.35: EXAMPLE OF ORGANIZATION OF ECONOMIC DATA 79
FIGURE 6.36: COUNT AND PERCENTAGE OF COMPANIES TRADED ON SPECIFIC EXCHANGES 81
FIGURE 6.37: AVERAGE YEARLY RETURNS OF COMPOSITE INDICES 81
FIGURE 6.38: AVERAGE RETURN 82
FIGURE 6.39: AVERAGE VOLATILITY 82
FIGURE 6.40: AVERAGE BETA 83
FIGURE 6.41: RELATIONSHIP BETWEEN RELATED PARTY TRANSACTIONS AND AVERAGE
REVENUE 85
FIGURE 6.42: RELATIONSHIP BETWEEN LEGAL PROCEEDINGS AND AVERAGE REVENUE 85
FIGURE 6.43: RELATIONSHIP BETWEEN AUDITORS OPINION AND AVERAGE REVENUE 85
FIGURE 6.44: RELATIONSHIP BETWEEN RETIREMENT PLAN AND AVERAGE REVENUE 85
FIGURE 6.45: RELATIONSHIP BETWEEN AUDITOR CHANGE AND AVERAGE CHANGE IN REVENUE
85
FIGURE 6.46: RELATIONSHIP BETWEEN TURNOVER OF MANAGEMENT AND AVERAGE CHANGE IN
REVENUE 85
FIGURE 7.1: TYPE I ERROR BY YEARS PRIOR TO BANKRUPTCY 92
FIGURE 8.1: DEA FRONTIER OF BEST PERFORMERS 95
FIGURE 8.2: DEA FRONTIER OF WORST PERFORMERS 95
FIGURE 8.3: LAYERING TECHNIQUE FOR NORMAL DEA 96
FIGURE 8.4: EXAMPLE OF METRICS AND SECOND STAGE ANALYSIS 96
FIGURE 8.5: EXAMPLE OF HANDLING NEGATIVE DATA 99
FIGURE 8.6: CLASSIFICATION WITH ONE CUT-OFF VALUE (X) 100
FIGURE 8.7: CLASSIFICATION WITH TWO CUT-OFF VALUES (X AND Y) 101
FIGURE 9.1: DISTRIBUTION OF SCORES GENERATED FROM PCA NORMAL MODEL 102
FIGURE 9.2: DISTRIBUTION OF SCORES GENERATED FROM PCA INVERSE MODEL 104
FIGURE 9.3: DISTRIBUTION OF SCORES GENERATED FROM LR NORMAL MODEL 105
FIGURE 9.4: SUMMARY OF NORMAL PROFITABILITY MODEL A 107
FIGURE 9.5: SUMMARY OF NORMAL PROFITABILITY MODEL B 107
FIGURE 9.6: SUMMARY OF INVERSE PROFITABILITY MODEL A 108
FIGURE 9.7: SUMMARY OF INVERSE PROFITABILITY MODEL B 108
FIGURE 9.8: EXAMPLE OF CONSISTENT NORMAL AND INVERSE MODELS 109
FIGURE 9.9: EXAMPLE OF INCONSISTENT NORMAL AND INVERSE MODELS 110
FIGURE 9.10: PARALLEL CLASSIFICATION 111
FIGURE 9.11: SERIAL CLASSIFICATION (NORMAL THEN INVERSE DEA) 112
FIGURE 9.12: SERIAL CLASSIFICATION (INVERSE THEN NORMAL DEA) 113
FIGURE 9.13: SUMMARY OF SOLVENCY WORST-PRACTICE MODEL C 114
FIGURE 10.1: SUMMARY OF INCOME STATEMENT (IS) METRIC 116
FIGURE 10.2: PREDICTION BY IS METRIC FROM ONE YEAR BACK 117
FIGURE 10.3: SUMMARY OF BALANCE SHEET ASSETS (BSA) METRIC 118
xv

FIGURE 10.4: PREDICTION BY BSA METRIC FROM ONE YEAR BACK 118
FIGURE 10.5: SUMMARY OF BALANCE SHEET LIABILITIES (BSL) METRIC 119
FIGURE 10.6: PREDICTION BY BSL METRIC FROM ONE YEAR BACK 119
FIGURE 10.7: CASH FLOWS IN PRODUCT LIFE CYCLE (STICKNEY ET AL., 2007) 120
FIGURE 10.8: SUMMARY OF MANAGERIAL DECISION-MAKING (MDM) METRIC 121
FIGURE 10.9: PREDICTION BY MDM METRIC FROM ONE YEAR BACK 121
FIGURE 10.10: THE EFFECT OF INFLATION ON IS SCORES 123
FIGURE 10.11: THE EFFECT OF INFLATION ON BSA SCORES 123
FIGURE 10.12: THE EFFECT OF INFLATION ON BSL SCORES 123
FIGURE 10.13 PREDICTION BY IS METRIC FROM ONE YEAR BACK AND BASED ON REAL DOLLARS
124
FIGURE 11.1: GENERAL ECONOMIC PERFORMANCE BY YEAR I 126
FIGURE 11.2: GENERAL ECONOMIC PERFORMANCE BY YEAR II 126
FIGURE 11.3: APPAREL INDUSTRY PERFORMANCE BY YEAR 127
FIGURE 11.4: HOUSING MARKET PERFORMANCE AND MONTHS FOR A SALE BY YEAR 128
FIGURE 11.5: HOUSING MARKET PERFORMANCE AND CONSTRUCTION BY YEAR 128
FIGURE 11.6: PRICES BY YEAR 129
FIGURE 11.7: GENERAL MARKET PERFORMANCE BY YEAR 129
FIGURE 11.8: ALL-ENCOMPASSING MARKET AND ECONOMIC INDICATOR BY YEAR 130
FIGURE 11.9: POSSIBILITIES OF ADDING MARKET AND ECONOMIC FACTORS INTO DEA MODELS
131
FIGURE 12.1: SECOND STAGE RESULTS WITH IS, BSA AND BSL 136
FIGURE 12.2: SECOND STAGE RESULTS WITH IS, BSA, BSL AND CFO 136
FIGURE 12.3: SECOND STAGE RESULTS WITH IS, BSA BSL AND MDM 136
FIGURE 12.4: SECOND STAGE RESULTS WITH IS, BSA, BSL, CFO AND MDM 136
FIGURE 12.5: CLASSIFICATION BY LAYER IN ABSOLUTE NUMBERS FROM ONE YEAR BACK 137
FIGURE 12.6: CLASSIFICATION BY LAYER IN PERCENTAGES FROM ONE YEAR BACK 138
FIGURE 12.7: PREDICTION BY SECOND STAGE MODEL FROM ONE YEAR BACK 138
FIGURE 12.8: ROC CURVES FOR ONE YEAR BACK 139
FIGURE 12.9: ROC CURVES FOR ONE YEAR BACK WITH CFO 140
FIGURE 12.10: TYPE I AND II ERRORS TRADE-OFF 141
FIGURE 12.11: CORRELATION BETWEEN LAYERING AND NON-LAYERING TECHNIQUES 143
FIGURE 12.12: AVERAGE SCORES BY YEAR 144
FIGURE 12.13: ALL COMPANIES (3.4-TO-1 RATIO) 146
FIGURE 12.14: 3-TO-1 RATIO 146
FIGURE 12.17: PROJECTIONS FOR METRICS IN THE SECOND STAGE FOR SELECTED DMUS 147
FIGURE 12.18: PROJECTIONS FOR IS METRIC VARIABLES FOR SELECTED DMUS 148
FIGURE 12.19: IS PROJECTIONS FOR COGS 149
FIGURE 12.20: IS PROJECTIONS FOR SGA 149
FIGURE 12.21: IS PROJECTIONS FOR NET INTEREST EXPENSE 149
FIGURE 12.22: IS PROJECTIONS FOR INCOME TAX EXPENSE 149
FIGURE 12.23: IS PROJECTIONS FOR REVENUE 150
FIGURE 12.24: IS PROJECTIONS FOR NET INCOME 150
FIGURE 12.25: BSA PROJECTIONS FOR MARKETABLE SECURITIES 150
FIGURE 12.26: BSA PROJECTIONS FOR ACCOUNTS RECEIVABLE 150
FIGURE 12.27: BSA PROJECTIONS FOR INVENTORIES 151
FIGURE 12.28: BSA PROJECTIONS FOR CURRENT ASSETS 151
xvi

FIGURE 12.29: BSA PROJECTIONS FOR LONG-TERM INVESTMENT IN SECURITIES 151
FIGURE 12.30: BSA PROJECTIONS FOR NET PPE 151
FIGURE 12.31: BSA PROJECTIONS FOR GOODWILL 152
FIGURE 12.32: BSA PROJECTIONS FOR TOTAL ASSETS 152
FIGURE 12.33: BSA PROJECTIONS FOR CASH 152
FIGURE 12.34: BSA PROJECTIONS FOR RETAINED EARNINGS 152
FIGURE 12.35: BSA PROJECTIONS FOR SHAREHOLDERS EQUITY 153
FIGURE 12.36: BSL PROJECTIONS FOR ACCOUNTS PAYABLE 153
FIGURE 12.37: BSL PROJECTIONS FOR NOTES PAYABLE AND SHORT-TERM DEBT 153
FIGURE 12.38: BSL PROJECTIONS FOR CURRENT MATURITIES OF LONG-TERM DEBT 154
FIGURE 12.39: BSL PROJECTIONS FOR CURRENT LIABILITIES 154
FIGURE 12.40: BSL PROJECTIONS FOR LONG-TERM DEBT 154
FIGURE 12.41: BSL PROJECTIONS FOR TOTAL LIABILITIES 154
FIGURE 12.42: BSL PROJECTIONS FOR RETAINED EARNINGS 155
FIGURE 12.43: BSL PROJECTIONS FOR SHAREHOLDERS EQUITY 155
FIGURE 12.44: MDM PROJECTIONS FOR RELATED PARTY TRANSACTIONS 155
FIGURE 12.45: MDM PROJECTIONS FOR AUDITORS OPINION 155
FIGURE 12.46: MDM PROJECTIONS FOR CHANGE OF AUDITOR 156
FIGURE 12.47: MDM PROJECTIONS FOR MANAGEMENT TURNOVER 156
FIGURE 12.48: MDM PROJECTIONS FOR LEGAL PROCEEDINGS 156
FIGURE 12.49: MDM PROJECTIONS FOR RETIREMENT PLAN 156
FIGURE 12.50: SECOND STAGE PROJECTIONS FOR IS METRIC 157
FIGURE 12.51: SECOND STAGE PROJECTIONS FOR BSA METRIC 157
FIGURE 12.52: SECOND STAGE PROJECTIONS FOR BSL METRIC 157
FIGURE 12.53: SECOND STAGE PROJECTIONS FOR MDM METRIC 157
FIGURE 13.1: DISTRIBUTION OF SECOND STAGE LAYERED SCORES 158
FIGURE 13.2: PROBABILITY OF BANKRUPTCY 160
FIGURE 13.3: TIME WINDOWS FOR MODEL TESTING 161
FIGURE 14.1: SUMMARY OF METHODOLOGY AS APPLICABLE TO ALL INDUSTRIES 167
FIGURE B.1: ONE INPUT AND ONE OUTPUT CASE 177
FIGURE B.2: TWO INPUTS AND ONE OUTPUT CASE 178
FIGURE B.3: ONE INPUT AND TWO OUTPUTS CASE 180
FIGURE F.1: DISTRIBUTION OF SCORES GENERATED FROM LR INVERSE DEA MODEL 189
FIGURE F.2: PREDICTION BY IS METRIC FROM TWO YEARS BACK 190
FIGURE F.3: PREDICTION BY IS METRIC FROM THREE YEARS BACK 190
FIGURE F.4: PREDICTION BY BSA METRIC FROM TWO YEARS BACK 191
FIGURE F.5: PREDICTION BY BSA METRIC FROM THREE YEARS BACK 191
FIGURE F.6: PREDICTION BY BSL METRIC FROM TWO YEARS BACK 192
FIGURE F.7: PREDICTION BY BSL METRIC FROM THREE YEARS BACK 192
FIGURE F.8: PREDICTION BY MDM METRIC FROM TWO YEARS BACK 193
FIGURE F.9: PREDICTION BY MDM METRIC FROM THREE YEARS BACK 193
FIGURE F.10: CLASSIFICATION BY LAYER IN ABSOLUTE NUMBERS FROM TWO YEARS BACK 194
FIGURE F.11: CLASSIFICATION BY LAYER IN PERCENTAGES FROM TWO YEARS BACK 194
FIGURE F.12: PREDICTION BY SECOND STAGE MODEL FROM TWO YEARS BACK 194
FIGURE F.13: CLASSIFICATION BY LAYER IN ABSOLUTE NUMBERS THREE YEARS BACK 195
FIGURE F.14: CLASSIFICATION BY LAYER IN PERCENTAGES FROM THREE YEARS BACK 195
FIGURE F.15: PREDICTION BY SECOND STAGE MODEL FROM THREE YEARS BACK 195
FIGURE F.16: IS PROJECTIONS FOR COGS 196
FIGURE F.17: IS PROJECTIONS FOR SGA 196
xvii

FIGURE F.18: IS PROJECTIONS FOR NET INTEREST EXPENSE 196
FIGURE F.19: IS PROJECTIONS FOR INCOME TAX EXPENSE 196
FIGURE F.20: IS PROJECTIONS FOR REVENUE 196
FIGURE F.21: IS PROJECTIONS FOR NET INCOME 196
FIGURE F.22: BSA PROJECTIONS FOR MARKETABLE SECURITIES 197
FIGURE F.23: BSA PROJECTIONS FOR ACCOUNTS RECEIVABLE 197
FIGURE F.24: BSA PROJECTIONS FOR INVENTORIES 197
FIGURE F.25: BSA PROJECTIONS FOR CURRENT ASSETS 197
FIGURE F.26: BSA PROJECTIONS FOR LONG-TERM INVESTMENTS IN SECURITIES 197
FIGURE F.27: BSA PROJECTIONS FOR NET PPE 197
FIGURE F.28: BSA PROJECTIONS FOR GOODWILL 198
FIGURE F.29: BSA PROJECTIONS FOR TOTAL ASSETS 198
FIGURE F.30: BSA PROJECTIONS FOR CASH 198
FIGURE F.31: BSA PROJECTIONS FOR RETAINED EARNINGS 198
FIGURE F.32: BSA PROJECTIONS FOR SHAREHOLDERS EQUITY 198
FIGURE F.33: BSL PROJECTIONS FOR ACCOUNTS PAYABLE 198
FIGURE F.34: BSL PROJECTIONS FOR NOTES PAYABLE AND SHORT-TERM DEBT 199
FIGURE F.35: BSL PROJECTIONS FOR CURRENT MATURITIES OF LONG-TERM DEBT 199
FIGURE F.36: BSL PROJECTIONS FOR CURRENT LIABILITIES 199
FIGURE F.37: BSL PROJECTIONS FOR LONG-TERM DEBT 199
FIGURE F.38: BSL PROJECTIONS FOR TOTAL LIABILITIES 199
FIGURE F.39: BSL PROJECTIONS FOR RETAINED EARNINGS 199
FIGURE F.40: BSL PROJECTIONS FOR SHAREHOLDERS EQUITY 200
FIGURE F.41: MDM PROJECTIONS FOR RELATED PARTY TRANSLATIONS 200
FIGURE F.42: MDM PROJECTIONS FOR AUDITORS OPINION 200
FIGURE F.43: MDM PROJECTIONS FOR CHANGE IN AUDITORS 200
FIGURE F.44: MDM PROJECTIONS FOR MANAGEMENT TURNOVER 200
FIGURE F.45: MDM PROJECTIONS FOR LEGAL PROCEEDINGS 200
FIGURE F.46: MDM PROJECTIONS FOR RETIREMENT PLANS 201
FIGURE F.47: SECOND STAGE PROJECTIONS FOR IS METRIC 201
FIGURE F.48: SECOND STAGE PROJECTIONS FOR BSA METRIC 201
FIGURE F.49: SECOND STAGE PROJECTIONS FOR BSL METRIC 201
FIGURE F.50: SECOND STAGE PROJECTIONS FOR MDM METRIC 201
FIGURE F.51: PREDICTION BY SECOND STAGE MODEL FROM ONE YEAR BACK, 1996-2008 202
FIGURE F.56: PROBABILITY OF BANKRUPTCY FROM ONE YEAR BACK, 1996-2009 204

xviii

List of Abbreviations

Acronym Definition Acronym Definition
AD Accumulated Depreciation MD&A Managements Discussion and Analysis
ADD Additive Model MDA Multiple Discriminant Analysis
AR Accounts Receivable ME Market and Economic
AP Accounts Payable MS Marketable Securities
BCC Banker-Charnes-Cooper Model Net IE Net Interest Expense
CA Current Assets Net PPE Net Property, Plant and Equipment
CC Change in Cash NI Net Income
CCR Charnes-Cooper-Rhodes Model Notes Notes to the Financial Statement
CFF Cash Flow From Financing Activities NPSTD Notes Payable and Short-Term Debt
CFI Cash Flow From Investing Activities OTC Over-the-Counter
CFO Cash Flow From Operating Activities Other CA Other Current Assets
COGS Cost of Goods Sold Other NonCA Other Non-Current Assets
CL Current Liabilities PCA Principal Component Analysis
CM Current Maturities of Long-Term Debt PM Profit Margin
DEA Data Envelopment Analysis PPE Property, Plant and Equipment
DMU Decision Making Unit RE Retained Earnings
EBIT Earnings Before Interest and Taxes Rev Revenue
FASB Financial Accounting Standards Board ROA Return on Assets
Goodwill Goodwill ROCE Return on Common Equity
IE Interest Expense RTS Returns-to-Scale
IFRS International Financial Reporting Standards SBM Slacks-based Measure of Efficiency Models
II Interest Income SE Shareholders Equity
Inv Inventories SEC Securities and Exchange Commission
ITE Income Tax Expense SGA Selling, General and Administrative Expenses
Lev Leverage TA Total Assets
LR Logistic Regression TAT Total Assets Turnover
LTD Long-Term Debt TL Total Liabilities
LTIS Long-Term Investment in Securities WC Working Capital

xix

List of Symbols

Symbol Definition
A Active, Non-bankrupt State
B

Bankrupt State
B-1

One Year Prior to Bankruptcy State
B-2

Two Years Prior to Bankruptcy State
B-3

Three Years Prior to Bankruptcy State
,
i
Beta, Beta of Firm
E(R
i
) Expected Return of Firm
E(R
m
) Expected Return of Market
R
i
Return of Firm
R
f
Risk-Free Rate of Interest
R
m
Return of Market
Set of Non-Negative Intensity Variables
m Total Number of Inputs
Efficiency (Output Oriented Model)
*
Optimal Efficiency (Output Oriented Model)
BCC
BCC Efficiency (Output Oriented Model)
N Total Number of Inputs and Outputs
SBM Efficiency or Spearmans Rank Correlation Coefficient
*

SBM Optimal Efficiency
Standard Deviation
s Total Number of Outputs
i
s ,
i
t

Input Slack for i
th
Input (Input and Output Oriented, respectively)
+
r
s ,
+
r
t
Output Slack for r
th
Output (Input and Output Oriented, respectively)
Efficiency (Input Oriented Model)
*
Optimal Efficiency
BCC
BCC Efficiency (Input Oriented Model)
u
*
Best Set of Output Weights
u
o
Unrestricted Dual Variable
u
r
Output Weight Given to r
th
Input
v Value Function
v
*
Best Set of Input Weights
v
o
Unrestricted Dual Variable
v
i
Input Weight Given to i
th
Input
x Input
x
ij
Amount of i
th
Input to DMU j
) , (
o o
y x

Coordinates of Virtual Linear Composite, Target
y Output
y
ij
Amount of r
th
Output to DMU j
Z Altmans Z-Score
DMU DEA Score

1

1 Introduction
Bankruptcy is a legally declared financial inability of an individual or organization to meet
their debt obligations. In the U.S., nearly 1.59 million filings for bankruptcy protection were
made in 2010, an increase of 8% from 2009 and 43% from 2008 (U.S. Courts, 2011). The
effect of bankruptcy is two-fold. Creditors (financial institutions, government financing
programs) lose interest payments and their principal investment while business owners are
subject to unpleasant legal, financial and personal consequences. For example, a trustee may
be assigned to the bankrupt business and force a sale of its property to pay back creditors
without the consent of the owner. This inevitably harms the owners reputation,
discouraging potential associates, damaging existing relationships and making it difficult to
obtain credit in the future.
Clearly, there are economic and social incentives for both parties to have the capacity to
predict the probability of bankruptcy. For creditors, accurate prediction would translate to
better assessment or pricing of interest payments with respect to risk. Essentially, financial
institutions and government financing programs could avoid making wrong credit decisions
by distinguishing companies in or heading for trouble (as to mitigate potential loss) from
those in good health (as to not refuse potential profits). For owners, prediction would
translate to prevention, buying time for them to secure more financing and avoid potential
demise.
Current mathematical models predict bankruptcy with common financial ratios as inputs,
disregarding market and economic factors. Also, while existing techniques make
dichotomous predictions (i.e. fail or survive) on the state of firms, they do not provide
strategies for owners to improve their operations when bankruptcy is imminent and finding
external sources of cash unlikely. Therefore, the objective of this research is to develop a
Data Envelopment Analysis (DEA) based model that assesses the likelihood of failure of
American retail-apparel companies and suggests preventative measures based on data
available from financial statements and their accompanying Notes (which provide hints on
managerial decision-making) as well as market and economic influences. It is hypothesized
that the inclusion of variables that reflect managerial decision-making, and market and
economic factors, enhance the predictive power of mathematical models that consider
2

financial data exclusively. Also, because this work is specific to retail-apparel, it is critical to
develop of a methodology that can be easily adapted for another industry.
This thesis is structured as follows:
Chapter 2 provides the basics on accounting and financial statement analysis, as well
as a brief literature survey on bankruptcy prediction and fraud.
Chapter 3 profiles the U.S. retail-apparel industry.
Chapter 4 offers a full background on Data Envelopment Analysis, its different
models and their mathematical formulations, with examples provided in the
Appendix.
Chapter 5 outlines the objectives of this research as motivated by the limitations of
the current techniques discussed in Chapters 2, 3 and 4.
Chapter 6 summarizes the collection, preparation and exploration of financial,
managerial decision-making, and market and economic data.
Chapter 7 describes the benchmarks of this work.
Chapter 8 outlines the methodology.
Chapters 9, 10, 11, 12, and 13 present and discuss the results.
Chapter 14 summarizes the major contributions of this thesis and lists
recommendations for future work.

3

2 Background I: Accounting and Financial Statement Analysis
2.1 The Annual Report
The U.S. Financial Accounting Standards Board (FASB) issues a common set of
authoritative rules, guides, conventions and procedures that all companies must adhere to in
recording their transactions and preparing their annual reports. Here, the most important
parts of the annual report the financial statements and its accompanying Notes, the
Managements Discussion and Analysis (MD&A) and the Auditors Report are described.

2.1.1 Balance Sheet
A balance sheet provides a snapshot of a companys financial resources and obligations at a
point in time. It indicates how money is distributed among generally recognized areas of a
business, namely, assets, liabilities and shareholders equity (Table 2.1).
Table 2.1: Common Entries in a Balance Sheet (Gill, 1994)
Assets: Anything that a company owns with monetary value. There are two types current assets and non-current
assets which are listed on a balance sheet in decreasing order of liquidity.
Current
Assets
All items that can be converted into cash quickly, in less than a year. e.g. cash and cash
equivalents, marketable securities, accounts receivable, inventories, prepaid expenses
Non-Current
Assets
Items that cannot be converted into cash quickly. e.g. property, plant and equipment (PPE),
accumulated depreciation, long-term investments in securities, amortizable intangible assets (i.e.
patents, trademarks, copyrights), goodwill and non-amortizable intangibles
Liabilities: Everything that a company owes. There are two types: current liabilities and non-current liabilities.
Current
Liabilities
The sum of monies owed that must be paid within 12 months. e.g. accounts payable, notes and
loans payable, short-term provisions, current maturities or interest on long-term debt, accruals (i.e.
taxes or salaries accumulated against current profits but not yet due to be paid)
Non-Current
Liabilities
Loans with obligations due more than one year from the date of issuance. e.g. long-term debt (i.e.
mortgage), deferred taxes, bonds payable
Shareholders Equity or Net Worth or Share Capital: The difference between total assets and total liabilities,
which encompasses minority interest in subsidiaries, treasury stock, preferred stock, common stock and dividends
paid, and retained earnings.

2.1.2 Income Statement
An income statement indicates how transactions (revenues and expenses, Table 2.2) are
transformed into net income or loss, over a period of time, measured either in months or a
year. The balance sheet and income statement are linked by Equation 2.1. Companies are
also required to disclose their earnings per share which reveals how much money
shareholders would receive for each share of stock they own if a firm distributed all of its net
income for the period (Equation 2.2).
4

Retained Earnings
year t
= Retained Earnings
year t-1
Dividends Paid
year t
+ Net Income
year t (2.1)
Earnings Per Share =
Net Income - Preferred Stock Dividends
Weighted Average Common Stock Shares Outstanding

(2.2)

Table 2.2: Common Entries in an Income Statement (Gill, 1994)
Operating Revenues
Revenues Money made from goods/services sold. The total dollar volume of all sales
minus returns, allowances, discounts and rebates.
Operating Expenses
Cost of Goods Sold Money spent to have goods/services ready which includes raw materials, direct
labour, delivery and overhead (spread over cost of units).
Selling, General &
Administrative Expenses
Non-production costs such as salaries, legal and professional fees, utilities,
insurance, rents, commissions and travel expenses, advertising, etc.
Depreciation and Amortization Decrease in value of assets purchased over time, reflecting wear and tear.
Other Operating Expenses Expenses or losses related to primary business operations (e.g. R&D).
Operating Income = Operating Revenue Operating Expenses
Non-Operating Gains Income generated from non-primary business activities (e.g. rent, patents) or
unusual or infrequent transactions (e.g. sale of securities or fixed assets)
Non-Operating Losses Expenses unrelated to primary operations (e.g. loss on sales of securities or fixed
assets, strikes).
Non-Operating Income = Non-Operating Revenue Non-Operating Expenses
Earnings Before Interest and Income Taxes (EBIT) = Operating Income + Non-Operating Income
Interest Income Money received from external investments either as dividends from another
companys earnings or interest made from lending money to others.
Interest Expense Expense of borrowing money.
Net Interest Expense = Interest Expense Interest Income
Earnings Before Income Taxes = EBIT Net Interest Expense
Earnings (Losses) from
Discontinued Operations
Excludes shifting business location, stopping production temporarily or changes
due to technological improvement.
Extraordinary Gains (Losses) Items that are both abnormal and infrequent (e.g. natural disasters, expropriation,
prohibitions under new regulations).
Changes in Accounting An example: deciding to depreciate an investment property that has not
previously been depreciated. Changes in estimates (i.e. useful life of fixed
assets) do not qualify.
Income Taxes Amount of taxes paid according to income.
Net Income = EBIT Net Interest Expense Earnings or Losses from Discontinued Operations, Extraordinary
Events, Changes in Accounting Income Taxes

2.1.3 Cash Flow Statement
The cash flow statement summarizes inflows and outflows of cash and cash equivalents of a
company during a specified period of time, and ties into the balance sheet (Figure 2.1).
Operating Cash Flows Current Assets Current Liabilities Operating Cash Flows
Investing Cash Flows
Capital (Fixed)
Assets
Non-Current Liabilities
} Financing Cash Flows
Shareholders Equity
Figure 2.1: Relationship between the Balance Sheet and Cash Flows
5

It is a cash-basis
1
report on three activities: operating, investing and financing (Table 2.3).
Table 2.3: Financial Activities Reported in a Cash Flow Statement
Operating Activities are related to the principle line of business.
Cash Inflows (Receipts) Cash Outflows (Payments)
Sales to or collection from customers
Receipts of interest and dividends from
investments

Payment to suppliers and contractors in purchase
of services or goods to resell
Payment to employees (salaries and wages)
Payment of income tax
Payment of interest on liabilities
Payment of rent and utilities
Investing Activities include capital expenditures and investments unrelated to the normal line of business.
Sale or disposal of capital assets (PPE)
Sale or maturity of investments that are not
cash equivalents
Collection of long-term loans receivables
Acquisition or purchase of capital assets (PPE)
Purchase of investments that are not cash
equivalents
Issuing long-term loans to others
Financing Activities reveal the companys debt or equity financing.
Borrowing money on notes, mortgages, bonds,
etc. from creditors
Issuing equity securities (stock, shares) to
owners
Repay principal amount of debts to creditors
(excluding interest)
Repurchase of equity securities from owners
Payment of dividends to owners

There are two methods for preparing the cash flow statement direct and indirect
both yielding the same result. While the direct method is more easily understood, the
indirect method is almost universally used (Epstein, 2007). The direct method reports major
classes of receipts and payments, and then arrives at the net cash effect. For example,
operating activities prepared under the direct method may resemble the following:
Cash Receipts from Customers
= Net Sales per Income Statement + Beginning Balance in Accounts Receivable
Ending Balance in Accounts Receivable
+ Cash Payments for Inventory
= Beginning Inventory Ending Inventory + Beginning Balance in Accounts
Payable Ending Balance in Accounts Payable
+ Cash Paid to Employees
= Salaries and Wages per Income statement + Beginning Balance in Salaries and
Wages Payable Ending Balance in Salaries and Wages Payable
+ Cash Paid for Operating Expenses
= Operating Expenses per Income Statement Depreciation Expenses Increase
or Decrease in Prepaid Expenses Decrease or Increase in Accrued Expenses
+ Taxes paid
= Tax Expense per Income Statement + Beginning Balance in Taxes Payable
Ending Balance in Taxes Payable
+ Interest paid
= Interest Expense per Income Statement + Beginning Balance in Interest Payable
Ending Balance in Interest Payable
= Net Cash Provided By (Used In) Operating Activities

1
In cash-basis accounting, revenues/expenses are recognized when cash is received/paid. In accrual-basis accounting,
revenue is recorded when it is earned and realized regardless of when monies are received while expenses match with
revenues, regardless of when they are actually paid.
6

Similar calculations can be made from the balance sheet to determine the cash flows to be
reported in the investing and financing activities sections of the cash flow statement.
Under the indirect method, net cash is derived from net income less the change in
working capital accounts and with adjustments in non-working capital accounts. Working
capital accounts include current assets and current liabilities, while non-working capital
accounts are those that are non-cash entries such as depreciation, deferred income taxes,
stock option expense and gain/loss on the disposition of property, plant and equipment
(PPE). The presentation of operating activities may resemble:
Net Income per Income Statement
entries to income accounts that do not represent cash flows
+ entries to expense accounts that do not represent cash flows
= Cash Flows Before Movements in Working Capital
the change in working capital
Increase in current assets would be negative because cash was spent or converted into other
current assets, thereby reducing the cash balance.
Decrease in current assets would be positive because they were converted into cash.
Increase in current liabilities would be positive since more liabilities imply less cash was spent.
Decrease in current liabilities would be negative because cash was spent to reduce liabilities
= Net Cash Provided By (Used In) Operating Activities

The cash flows from investing and financing activities would be presented as under the direct
method (Epstein, 2007).

2.1.4 Notes to the Financial Statements
Notes to the Financial Statements (herein referred to as Notes) provide supplementary
information to the financial statements. They are not presented in any specific format but
generally explain the accounting method and assumptions used to prepare the statements, the
computation of items in the financial statements, and the valuations for how particular
accounts are represented. Furthermore, they offer a comprehensive assessment of a
companys financial situation, particularly on outstanding debt, going concern, contingent
liabilities or context (lawsuits, mergers and acquisitions), which clarify the financial
numbers. Some significant accounting policies in the Notes are:
Asset Types and Method of Valuation, Depreciation and Amortization: A
company must describe how assets are valued and used because the selection of both
valuation and depreciation methods may greatly impact net income. For example, a
slower depreciation rate translates to less expenses and higher net income.
7

Recognition of Revenues and Expenses: The timing of revenue and expense
recognition affects the profit reported. Revenue recognition depends on the certainty
of collectability, which relies on the business environment, customers financial
condition, historical collection experience, and the aging of accounts receivable.
Expense recognition may depend on whether product development is in-house or
outsourced, and/or when advertising expenses are written-off (at the time of print or
air versus over a period of time as with catalogue sales).
Commitments: A companys debt structure is divided into short-term and long-term
borrowings, and income tax obligations. Of particular interest are leases. A capital
lease provides ownership at no cost or at a low residual value at the end of the lease;
it appears as a long-term obligation on the balance sheet since the lessee bears the
risk. An operating lease offers no ownership; it is mentioned in the Notes but
excluded from the balance sheet since the lessor bears the risk. In some cases, the
capitalization of operating leases is necessary to reflect the firms debt structure more
accurately. A lease is a capital lease if it meets any of one four conditions (FASB,
1976):
1. If it extends for at least 75% of the assets total expected useful life (i.e. the
lessee uses the asset for most of its life).
2. If it transfers ownership to the lessee at the end of the lease term.
3. If it seems likely that the lessor will transfer ownership to the lessee because
of a bargain purchase option that gives the lessee the right to purchase the
asset for a price less than the expected fair market value of the asset when the
lessee exercises its option.
4. The present value of the contractual minimum lease payments equals or
exceeds 90% of the fair market value of the asset at the time of signing.
Pensions: For companies that offer pension benefits to their current and future
retirees, this obligation may be greater than that to creditors and shareholders. A
company with a defined benefit plan promises retirement benefits to each employee
whereas one with a defined contribution plan requires both employer and employee to
make contributions until retirement.
8

Stock-Based Compensation: If relevant, a company will describe its employee
incentive plans which may involve stock ownership.
Significant Relationships and Events: Significant relationships and events that may
impact financial positions include related-party transactions, legal proceedings,
restructuring, discontinued operations and contingencies. A company should discuss
the impact of a relevant relationship or event on the current year as well as on
expected future performance.

2.1.5 Managements Discussion and Analysis (MD&A)
Valuable information may be found in the Management's Discussion and Analysis (MD&A)
which provides an overview of a corporation noting any special or unusual circumstances
that may have affected the financial results over the past year. In the U.S., the MD&A is
monitored by the Securities and Exchange Commission (SEC) to ensure a company presents
all critical information about current operations, capital and liquidity. With respect to
operations, the MD&A should focus on: revenue and expense recognition; key profit results
and their consistency with previous years projections; its outlook on future financials based
on historical sales, product performance and improvements, research and development, and
economic or market conditions. With respect to capital, the MD&A should cover key points
affecting cash flow: any merger, acquisition or expansion that translates to major expenses
in the past year or carry over to future years; and current plans for new debt. With respect to
liquidity, the MD&A should discuss the firms cash position and ability to pay its creditors.
And finally, if applicable, the MD&A should touch upon issues such as restructuring charges,
impairments to assets, environmental and product liabilities, and allowance for doubtful
accounts.

2.1.6 Auditors Report
For credibility, a company must hire an independent accounting firm to audit their internal
controls and financial statements. Auditors raise any red flags about its financial results but
do not provide 100% assurance nor endorse the companys financial position. Their report
consists of three paragraphs:
9

1. Introduction outlines the time period that the audit covers and who is
responsible for the statement (to alleviate the auditors of their responsibility for
possible inaccuracies).
2. Scope describes how the audit is carried out and any material misstatements (or
errors) that significantly impact companys financial position.
3. Opinion states whether the financials are prepared in conformity with the
FASB.
A standard report is an unqualified opinion. A nonstandard report requires auditors to
explain their qualified opinion, which typically arises when there is a change in accounting
policies, debt-agreement violations, lawsuits, loss of major customers or market share, and
going-concern problems.

2.2 Financial Statement Analysis: Ratios
The analysis of a companys financial statements to determine corporate health is heavily
based on ratios. Financial ratios quantify specific aspects of a business, and enable
comparisons of strengths and weaknesses between firms within a given industry, against the
average and across different time periods. However, ratios of firms in different industries
facing different risks, capital requirements and competition, are not comparable and are one-
dimensional (Gill, 1994).
The following sections elaborate on common financial statement analyses.

2.2.1 Industry Analysis
Industry analysis examines the degree of competition and the degree of capital investment in
terms of return on assets (ROA), profit margin (PM) and total assets turnover (TAT) (Table
2.4). A company that requires a high capital investment is likely in an industry that is
monopolistic by nature (Type A). Consequently, ROA depends on high profit margin as
opposed to high assets turnover. In contrast, a company in an industry without high barriers
to entry faces competition. In this case, ROA is driven by efficiency (high asset turnover) as
opposed to high profit margin (Type C). It is noted that there is no one better way for a
company to operate: industries are simply different and as such, companies can only be
compared if they operate under the same conditions; thus, Type A firms are not comparable
10

to Type C firms. Also, in plotting these variables of a company over time, one can observe
whether its strategic focus is consistent.
Table 2.4: Industry Analysis (Stickney et al., 2007)
Type Competition
Level
Capital
Investment
Profit
Margin
Assets
Turnover
Strategic Focus
for ROA
A Low High Large Low Profit Margin
B Medium Medium Medium Medium Profit Margin and/or Assets Turnover
C High Low Small High Assets Turnover

2.2.2 Profitability Analysis
Mathematically, ROA is a product of profitability and efficiency, and can be broken down
into 3 levels.
Level I Return on Assets (ROA) is calculated by:
ROA =
Net Income + Interest 1 tax rate + Minority Interest in Net Income
Average Total Assets
(2.3)

ROA can be expressed as function of profit margin and total assets turnover,
introducing a second level:
ROA = Profit Margin for ROA Total Assets Turnover
(2.4)

Level II Profitability and Efficiency are defined as:
Profit Margin
=
Sales

(2.5)
Total Assets Turnover =
Sales
(2.6)

Both these measures can be examined more specifically, presenting a third level of
analysis.
Level IIIa Profit Margin can be related to different items on the income statement, as a
percentage of revenues, for example:
Cost of Goods Sold % =
Cost of Goods Sold
Sales
(2.7)
Selling, General and Administrative Expenses %
=
Selling, General and Administrative Expenses
Sales

(2.8)
Income Tax Expense % =
Income Tax Expense + Tax Rate Interest Expense
Sales
(2.9)

11

Level IIIb Total Assets Turnover can be studied in terms of accounts receivable,
inventory and fixed assets:
Accounts Receivable Turnover =
Sales
Average Accounts Receivable
(2.10)
Inventory Turnover =
Cost of Goods Sold
Average Inventory
(2.11)
Fixed Assets Turnover =
Sales
Average Fixed Assets
(2.12)

In addition to ROA, a companys Return on Common Equity (ROCE, Equation
2.13) reveals information about its capital structure leverage. If ROCE is greater than ROA,
then the company is positively leveraged; that is, it manages its debt well. If ROCE is equal
to ROA, then it is neutrally leveraged. If ROCE is less than ROA, then it is negatively
leveraged a signal for concern.
ROCE =
Net Income
Average Shareholders' Equity

=
Sales

Sales


(2.13)

Another common measure of profitability is return on sales:
Return on Sales = Operating Margin =
Operating Income
Revenue
(2.14)

2.2.3 Short-Term Liquidity and Long-Term Solvency
Liquidity refers to the ability of a firm to quickly and easily convert non-cash assets into cash
without incurring a significant loss. Common tests of liquidity focus on the relationship
between current assets and current liabilities, measuring the availability of cash to pay debt
(Table 2.5). Solvency refers to the ability to meet maturing obligations as they come due.
Common tests of solvency measure a companys capacity to repay long-term debt or
financial leverage (Table 2.5).

12

Table 2.5: Common Liquidity and Solvency Ratios
Short-Term Liquidity Ratios
Current Ratio
Current Assets
Current Liabilities

Quick Ratio
Quick Assets
2
Current Liabilities

Operating Cash Flow to Current Liabilities Ratio
Cash Flow from Operations
Average Current Liabilities

Days Receivable Outstanding (Average Collection Period)
365
Accounts Receivable Turnover

Days Inventory Outstanding (Average Inventory Period)
365
Inventory Turnover

Accounts Payable Turnover
Purchases
Average Accounts Payable

Days Payable Outstanding (Average Payment Period)
365
Accounts Payable Turnover

Revenues to Cash Ratio
Revenues
Average Cash Balance

Days Revenues in Cash
365
Revenues to Cash Ratio

Long-Term Solvency Ratios
Times Interest Earned (Interest Coverage) Ratio
Earnings Before Interest and Tax
Interest Expense

Cash Coverage Ratio
Operating Cash Flow Before Interest Tax
Interest Paid

Liabilities to Assets Ratio
Total Liabilities
Total Assets

Liabilities to Shareholders Equity (Debt to Equity) Ratio
Total Liabilities
Total Shareholders' Equity

Long-Term Debt to Long-Term Capital Ratio
Long-Term Debt
Long-Term Debt+ Total Shareholders' Equity

Long-Term Debt to Shareholders Equity Ratio
Long-Term Debt

2.3 Bankruptcy
Bankruptcy is a legally declared financial inability of an organization to meet its debt
obligations. Causes of bankruptcy include: financial mismanagement, poor location,
competition, difficulties with cash flow, loss of capitalization, high debt, lack of planning, tax
burdens and regulations, loss of a key person, lack of technology, poor record keeping,
personal issues, and natural disaster or accident leading to high insurance (Bradley et al.,
2004).

2
Quick Assets = cash + accounts receivable + notes receivable + short term investments + marketable securities

13

Bankruptcy is often incorrectly used as a synonym for insolvency: insolvency refers to a
financial state whereas bankruptcy is a matter of law. Insolvency occurs when either:
a companys liabilities exceed their assets (known as balance-sheet insolvency); or
a company cannot meet its debt obligations as they become due (known as cash flow
insolvency).
On the other hand, bankruptcy is a successful legal procedure that results from:
an application to the relevant court by a company in order to have itself declared
bankrupt; or
an application to the relevant court by a creditor of a company in order to have the
company declared bankrupt.
A state of insolvency can lead to bankruptcy but the condition may also be temporary and
fixable without legal protection from creditors. In other words, insolvency does not
necessarily lead to bankruptcy, but all bankrupt firms are considered insolvent.
Researchers have worked diligently to identify keys to business survival by applying
various factors from financial ratios derived from financial statements, to geographic
location, type of industry and competition, size and age to different mathematical
techniques. To date, there are over 200 journal articles reporting on bankruptcy prediction
using common financial ratios alone. In 2006, Ravi Kumar et al. conducted an extensive
review of nearly 120 papers, grouping them by methodology: 1. statistical techniques, 2.
neural networks, 3. case-based reasoning, 4. decision trees, 5. operational research, 6.
evolutionary approaches, 7. rough set based techniques, 8. other techniques subsuming fuzzy
logic, support vector machine and isotonic separation and 9. soft computing subsuming
seamless hybridization of all the above-mentioned techniques. A brief review will be
provided on two categories, relevant to this study:
1. Statistical techniques (Sections 2.3.1 and 2.3.2): traditionally used and serve as
benchmarks for this thesis; and,
2. Operations research (Section 4.9): the foundation of the approach taken in this work.

2.3.1 Altmans Bankruptcy Model
In 1968, Edward Altman developed the Z-Score bankruptcy predictor to gauge a firms
financial health (Table 2.6). Applying common financial ratios to multiple discriminant
14

analysis (MDA), the original Z-Score forecasts the probability of a manufacturing company
entering bankruptcy within a two-year period. There are 3 zones of discrimination: the safe
zone (Z 3), grey zone (1.8 < Z < 3) and distress zone (Z 1.8) (Altman, 1968). From a
dataset of 33 bankrupt and 33 non-bankrupt firms, the reliability of Altmans model is 72-
80% with Type I and II errors
3
presented in Table 2.7). Table 2.6 also provides the Z-Scores
for private and non-manufacturing firms which differ from the original by coefficients.
Table 2.6: Z-Scores by Firm Type and Zones of Discrimination (Altman, 1968; Hanson, 2003)

Original Z-Score for Public Manufacturing Firms
Assets Total
Sales
99 . 0
Debt Total of Value Book
Equity of Value Market
6 . 0
Assets Total
Taxes and Interest Before Earnings
3 . 3
Assets Total
Earnings Retained
4 . 1
Assets Total
Capital Working
2 . 1
+ +
+ + = Z

Zones of Discrimination: Z 3: Safe Zone; 1.8 < Z < 3: Grey Zone; Z 1.8: Distress Zone

Z'-Score for Private Firms
Assets Total
Sales
998 . 0
s Liabilitie Total of Value Market
420 . 0
Assets Total
107 . 3
Assets Total
Earnings Retained
847 . 0
Assets Total
Capital Working
717 . 0
+ +
+ + = Z

Zones of Discrimination: Z' 2.9: Safe Zone; 1.23 < Z' < 2. 9: Grey Zone; Z' 1.23: Distress Zone

Z'-Score for Public Non-Manufacturing Firms
s Liabilitie Total of Value Market
05 . 1
Assets Total
72 . 6
Assets Total
Earnings Retained
26 . 3
Assets Total
Capital Working
56 . 6
+
+ + = Z

Zones of Discrimination: Z' 2.6: Safe Zone; 1.1 < Z' < 2. 6: Grey Zone; Z' 1.1: Distress Zone

Table 2.7: Error Classification for Altmans Original Model
Year 1 Year 2
Type I 6% 28%
Type II 3% 6%
Total 5% 17%

The strengths of MDA are that it incorporates multiple financial ratios
simultaneously; it provides coefficients for the combined independent variables; and it is
simple to apply once the formula is derived. However, MDA can only select from ratios
provided to it which may not necessarily have the strongest explanatory power. Also, the
value of the cut-off score(s) that best distinguishes bankrupt from non-bankrupt firms is

3
A Type I Error misclassifies bankrupt firms whereas a Type II Error misclassifies non-bankrupt firms. Hence, a Type I
Error is more costly than a Type II Error.
15

subjective. Other limitations include the assumption of normal distribution of financial ratios
and the requirement that the covariance matrix of the explanatory variables be the same for
bankrupt and non-bankrupt firms.

2.3.2 Ohlsons Bankruptcy Model
In 1980, James Ohlson used maximum likelihood estimation of the so-called conditional
logit model to predict bankruptcy. The population boundaries of his dataset were restricted
by:
1. the period from 1970 to 1976;
2. the equity of each company having had to be traded on some stock exchange; and,
3. the classification of each company being industrial (no small or private corporations,
utilities, transport and financial services).
From a sample of 105 bankrupt and 2068 non-bankrupt firms, Ohlson derived the probability
of bankruptcy to be 1/(1+e
-y
), where
y = 1.32 0.407 SIZE + 6.03 TLTA 1.43 WCTA + 0.0757 CLCA 2.37
NITA 1.83 FUTL + 0.285 INTWO 1.72 ONNEG 0.521 CHIN
and SIZE = ln
Absolute Total Assets
GNP Implicit Price Deflator

TLTA =
Total Liabilities
Total Assets

WCTA =
Net Working Capital
Total Assets

CLCA =
Current Liabilities
Current Assets

NITA =
Net Income
Total Assets

FUTL =
Operating Working Capital
Total Liabilities

INTWO = one if net income was negative for the last two years; zero otherwise
OENEG = one if total liabilities exceeds total assets; zero otherwise
CHIN =
Net Income
t
-Net Income
t-1
|Net Income
t
|+|Net Income
t-1
|

A probability greater than 4% is concerning. Compared to Altman, Ohlson has
greater accuracy (Table 2.8). However, there is an important difference between the models
16

in that the magnitude of total assets relative to other variables is significant to bankruptcy
risk: Altman predicted that bankruptcy is imminent when the firms use of assets is
inefficient whereas Ohlson predicted that bankruptcy is unlikely due to the underlying
assumption that their absolute assets (large size) could be sold.
Table 2.8: Error Classification for Ohlsons Model
Year 1 Year 2 Year 3
Type I 12% 20% 24%
Type II 7% 18% 20%
Total 10% 19% 22%

These studies and those by pioneers Beaver (1966), Wilcox (1971), Deakin (1972),
Edmister (1972), Blum (1974), Libby (1975) and Moyer (1977) who used various statistical
methods (e.g. linear discriminant analysis, MDA, quadratic discriminant analysis, logistic
regression, factor analysis, maximum likelihood estimation) paved the way for others to
consider mathematical alternatives to bankruptcy prediction (refer to Ravi Kumar et al.
(2006)). These works also share some key methodological issues in bankruptcy prediction
research.
1. Sample sizes of bankrupt and non-bankrupt firms: The proportion of bankrupt firms
in the economy is substantially smaller than that of non-bankrupt firms. It is
inadvisable to make the groups equal (as Altman did) because this results in over-
fitting models towards the characteristic of bankrupt firms. Thus, the challenge is to
find a proportion of non-bankrupt firms that more closely reflects the populations of
firms but does not inherently bias the model.
2. Difference in company size and industry characteristics: Companies in different
industries have different structures and environments. Also, small firms may
experience greater difficulty obtaining funds than larger firms. Thus, models should
be developed for a specific industry and normalized for size.
3. Stability in bankruptcy prediction models over time. Bankruptcy laws and their
judicial interpretations, and frequency of filings due to economic change, evolve over
time. Hence, models should be periodically updated to reflect new financing vehicles
that emerge that help finance firms that were not originally available.
4. External influences. Companies are subject to market and economic risk factors.

17

2.4 Fraud
Financial statement fraud is usually committed to meet financial performance goals, analysts
forecasts, and stakeholders expectations, or to mask volatility and financial distress
(Ramamoorti, 2007). In a study of 204 organizations, Albrecht (2008) found that some
common characteristics include:
the CEO being the primary perpetrator (in 72% of fraud cases);
the last audit opinion prior to fraud containing going concerns, litigation and
uncertainties (24%);
the average financial statement fraud lasting 23.7 months;
the mean cumulative financial statement misstatement being $25M;
78% of fraudulent firms being listed on NASDAQ or OTC markets, 15% on NYSE,
and 7% on American Stock Exchange;
the industries with the most frauds being hardware/software (12%), financial services
providers (11%), health care providers (9%) and retailers/wholesalers (7%);
family relationships between directors and officers being fairly common (40%); and,
many experiencing net losses or being close to break-even positions in periods prior
to fraud.
Clearly, there is value in studying the characteristics of fraudulent reporting as a precursor to
bankruptcy. In general, there are two fraud detection methods based on: 1. financial
measures (quantitative data extracted from financial statements) and 2. non-financial
measures (either qualitative or quantitative data derived from other sections of the annual
report).

2.4.1 Investigating Fraud with Financial Measures
Earnings manipulation is defined as an instance in which a companys management
intentionally represents its financial performance more favourably than it really is. In 1999,
Beneish developed a model to distinguish manipulated from non-manipulated reporting by
using financial statement data to construct variables that would capture the effects of
manipulation and preconditions that might prompt companies to engage in such activity. His
sample consisted of 74 firms: 45% were in the manufacturing or services industries while
18

only 4 companies were in software, 3 companies were in computers, and 3 companies were
in audiovisual retail stores.
His probability of manipulation was given by the value of the probability density
function or the cumulative distribution function for the normal distribution with a given mean
and standard deviation at Y, where
Y = 4.840 + 0.92 DSRI + 0.528 GMI + 0.404 AQI + 0.892 SGI + 0.115
DEPI 0.172 SGAI 0.327 LVGI + 4.679 TATA
and DSRI =
Accounts Receivables
t
/Sales
t
t-1
/Sales
t-1

GMI =
Gross Margin
t-1
Gross Margin
t

AQI =
(1 - Current Assets + Net PP&E)
t
/Total Assets
t

(1 - Current Assets + Net PP&E)
t-1
/Total Assets
t-1

SGI =
Sales
t
Sales
t-1

DEPI =
Depreciation
t-1
/(Depreciation + PP&E)
t-1
Depreciation
t
/(Depreciation + PP&E)
t

SGAI =
SG&A
t
/Sales
t
SG&A
t-1
/Sales
t-1

LVGI =
(Long Term Debt + Current Liabilities)
t
/Total Assets
t

(Long Term Debt + Current Liabilities)
t-1
/Total Assets
t-1

TATA =
Net Income from Continuing Operations
t
- Cash Flow from Operations
t
Total Assets
t

Probabilities greater than 3% raised suspicion for earnings manipulation at which point,
any of the above ratios with a value greater than 1 should be analyzed.
Beneishs model was modified by others like Weidman (1999) as follows:
Manipulation Index = 2.224 + 0.221 DSRI + 0.102 GMI + 0.007 AQI +
0.062 DEPI + 0.198 SGAI 0.684 Abnormal Index + 0.001 Time Listed +
0.587 LVGI + 0.421 Positive Accruals Dummy 0.431 Declining Cash Sales
Dummy
where
19

Abnormal Index is computed by subtracting the difference in average holding
period (buy-and-hold) returns of issuing and non-issuing size-matched value-
weighted firms from the companys buy-and-hold return;
Time Listed is the number of months between the fiscal year end and the date
that the company was first listed on either the NYSE or NASDAQ;
Positive Accruals Dummy is equal to 1 if total accurals are positive in the
current and prior year, and 0 otherwise; and,
Declining Cash Sales Dummy is equal to 1 if cash sales in the current year are
lower than in the previous year, and 0 otherwise (n.b. Cash Sales
t

= Sales
t

Receivables).
Here, a probability greater than 11.72% indicated a very significant danger of
manipulation; between 5.99% and 11.72% indicated a serious risk of manipulation;
between 2.94% and 5.99% indicated a grey zone; and less than 2.94% indicated no
evidence of manipulation.
In another example, Kaminski et al. (2004) examined 21 financial ratios on 79
matched pairs of firms, spanning a time period of three years prior to and post the fraud year.
Five ratios were found to be significant prior to the fraud year (
Total Assets
,
Cost of Goods Sold
Sales
,
Fixed Assets
Total Assets
,
Interest Expense
Total Liabilities
and
Sales
) and when coupled with quadratic
discriminant analysis, 24-59% of fraudulent firms were correctly classified while the
accuracy of classifying non-fraudulent firms was 98%.
In addition to quadratic discriminant analysis, other data mining techniques have
been applied. For instance, Kirkos et al. (2007) studied 76 Greek manufacturing firms (half
fraud, half non-fraud) and tested 10 financial ratios
4
as variables in decision trees, Bayesian
belief networks and neural networks. The results were consistent: for the decision tree,
Altmans Z-Score was the first node, followed by NPTA, EBIT, COSAL and DEBTQ; for
the Bayesian belief, the significant ratios were Z-Score, NPTA, DEBTEQ, SALTA and
WCTA; and for the neural network, classification was claimed to be 100% correct when all
ratios were considered.

4
DEBTEQ = debt to equity ratio; SALTA = sales to total assets ratio; COSAL = sales minus gross margin; EBIT = earnings
before interest and taxes; WCAP = working capital; Z-Score = Altmans Z-Score; TDTA = total debt to total assets ratio;
NPTA = net profit to total assets ratio; WCTA = working capital to total assets ratio; GPTA = gross profit to total assets
ratio
20

2.4.2 Investigating Fraud with Non-Financial Data
Detecting fraud with non-financial data is common practice among professional auditors and
accountants. The American Institute of Certified Public Accountants (AICPA) has audit and
attest standards for conducting, planning and reporting such as the Statements on Auditing
Standards (SASs). In 1988, the Auditing Standards Board (ASB) of AICPA issued SAS No.
52 The Auditors Responsibility to Detect and Report Errors and Irregularities which was
later replaced by SAS No. 82 Consideration of Fraud in a Financial Statement Audit. SAS
No. 82 identifies 25 fraud risk factors in 3 categories: 1. management characteristics and
influence over control environment; 2. industry conditions; and, 3. operating and financial
stability characteristics (Table 2.9). It has served as guidelines for researchers aiming to
determine the probability of fraudulent reporting.
Table 2.9: SAS No. 82 (Apostolou, 2001)
Managements Characteristics and Influence over the Control Environment
Significant compensation tied to aggressive accounting practices
Managements failure to display appropriate attitude about internal control
Known history of securities law violations
Non-financial managements influence over GAAP principles or estimates
Strained management/auditor relationship
High turnover of senior management
Industry Conditions
High degree of competition or market saturation and declining margins
Rapid changes in industry and vulnerability to changing technology and product obsolescence
Company in declining industry
Effect of new accounting requirements on financial stability and profitability
Operating Characteristics and Financial Stability
Presence of aggressive incentive programs
Potential adverse consequences of poor financial results
Poor or deteriorating financial position with management guarantee of firms debt
Negative operating cash flow but reported earnings
Significant related-party transactions
Bank accounts or operations in tax-haven jurisdictions
Substance over form questions
Significant accounts based on estimates
Threat of imminent bankruptcy
Significant pressure to obtain capital
Unusually rapid growth or profitability relative to the industry
Difficulty in determining organizational control
Overly complex organization
Unusually high dependence on debt
High vulnerability to interest rates

In 2001, Apostolou et al. evaluated the relative importance of the SAS No. 82 risk
factors using Analytic Hierarchy Process: experts were first asked to indicate which category
was most important, then to score each risk factor within each category on a nine-point scale.
21

Managements characteristics were rated most important (58.5%), followed by operating and
financial stability (27.0%) and industry conditions (14.5%). The two most important
individual factors were significant compensation tied to aggressive accounting practices
(18.0%) and managements failure to display an appropriate attitude about internal controls
(16.5%). While Apostolous methodology was quite subjective, it provided insight into what
the average expert valued.
In a similar work, Bell et al. (2000) studied 77 fraud and 305 non-fraud engagements
based on 46 risk factors, 21 of which were derived from SAS No. 53. Auditors were asked
(yes or no) whether they believed a given risk factor was a reliable fraudulent reporting
predictor. Although many of the 25 non-SAS No. 53 risk factors lacked any relationship to
fraud (likely because of their infrequency), several SAS No. 53 factors did significantly
correlate with fraud:
management lied to the auditor or was overly evasive when responding to audit
inquiries (0.48)
weak internal control environment (0.46)
management's attitude unduly aggressive (0.42)
management places undue emphasis on meeting earnings projections (0.41)
significant difficult-to-audit transactions or balances are present (0.40)
Like Apostolou et al., Bell et al. found that factors related to managements characteristics
and influence over control environment were stronger indicators for fraud compared to those
related to industry conditions and operating characteristics and financial stability. This was
consistent with most studies like Smith et al. (2005) who also concluded that managements
failure to display an appropriate attitude about internal control was most important.

2.4.3 Critical Red Flags
Unlike the models in Section 2.4.1, which provided clear tangible measures of manipulation,
how does one determine if management has failed to display an appropriate attitude about
internal control or assess other SAS No. 82 risk factors that were vaguely described? Most
studies discussed in Section 2.4.2 were ambiguous with respect to data collection and
specification. Here, some strategies in financial reporting that management employ to
22

mislead stakeholders are listed, and techniques that bring these manipulations to light are
offered (Albrecht, 2008; Elitzur, 2009; Seow, 2009).
Manipulation of Earnings: There are three common approaches:
o Income-increasing to appear more profitable for greater bonuses to be paid
to executives;
o Income-decreasing to thwart industry-specific or antitrust actions against the
firm; and,
o Income smoothing to reflect economic results as management wishes them
to look, through redistribution of income statement credits and charges among
periods.
Improper or Premature Revenue Recognition: This strategy maximizes revenues
to influence market valuation even if it does not affect net income. A firm can
recognize revenues by either the percent completion method (PCM) which spreads
recognition over the entire contract period, or the completed contract method (CCM).
Quality of Income: High-quality earnings have a high correlation with cash flows,
which are more difficult to manipulate than reported income. Hence, comparing cash
flows with net income using Equation 2.15 or a plot of these variables over time
should reveal the level of accounting accruals used.
Quality of Income =
Operating Cash Flow
Net Income

(2.15)
A ratio of over 1 (i.e. operating cash flow greater than net income) suggests high
quality of income whereas a ratio of less than 1 is problematic. Moreover, if
operating cash flow is negative and net income is positive, the ratio serves as a red
flag, indicating a poor earnings quality.
It may also be worthwhile to run a regression (Equation 2.16) between the two
variables to observe the lag in converting net income (NI) into cash (CFO).
CFO
t+1
= + NI
t
(2.16)
From the regression, a positive , a high value for the adjusted R
2
(reflecting
variability of net income) and a p-value tending to zero indicate high quality of
numbers. Furthermore, the Durbin-Watson statistic reveals whether the regression
23

must be adjusted to account for seasonality. If such is the case, dummy variables are
added to Equation 2.16.
Off-Balance Sheet Liabilities: A root cause of many manipulation cases, off-
balance sheet liabilities improve the reported capital structure of the company by not
reporting some debt on the balance sheet. Liabilities may be disguised with the non-
capitalization of operating leases or the creation of special purpose entities (SPEs).
Notes pertaining to long-term debt and pensions should also be analyzed for whether
cash is available to pay these obligations.
Valuation of Assets and Liabilities: Because of its potential for accounting
creativity, the valuation of assets and liabilities must be justified. Overvaluing assets
may suggest that a firm believes to own more than it does, or the recording of
fictitious assets or the improper asset depreciation or amortization. Undervaluing
liabilities may indicate that a company owes less than it does. It may create excess
reserves by initially over-accruing a liability in one accounting period, and then
reducing the excess reserves to operating income in later accounting period.
Changes in Accounting Policies: It is important to understand the motivation behind
any changes and the impact on the financial numbers, especially when comparing
year-to-year results.
Change of Auditors: This change may signal disagreements on the accounting rules
between the management of the company and its auditors.
Auditors Opinion and Reputation: A nonstandard (qualified) opinion is a
disclaimer on accounting irregularities and encourages additional research through
the Notes and MD&A.
Management Turnover: When executive managers (especially key accounting and
financial personnel) leave the firm, it may be the result of an accounting problem in
the company.
Financial Distress: If liquidity, solvency and/or bankruptcy analysis reveal
difficulties for a firms ability to pay debt, management may be tempted to
manipulate their financial statements.
24

Growth of Receivables vs. Sales Growth: When a companys receivables
percentage growth exceeds sales growth, this may suggest sales are increased through
fictitious receivables, or that the firm prematurely recognizes revenues.
Executive Compensation: Studying the firms executive compensation plans may
reveal any incentive to manipulate earnings (i.e. if bonuses are a percentage of profit).
Trading by Insiders: Stock transactions by insiders could indicate a future trend in
earnings. For example, if many insiders sell their shares, it may be in anticipation of
earning problems.
Related-Party or Enormous Transactions with a Small Number of Customers:
Transactions between the firm and its related parties (i.e. officers, directors,
employees, affiliates, subsidiaries and SPEs) are often at the heart of accounting
irregularities. Also, a drastic increase in sales due to a small number of transactions
with customers may be an issue in revenue recognition.
Labelling: Companies may label unusual accounts in their financial statements
differently to manipulate their earnings.
Opportunistic Use of Exogenous Events: Some companies may emphasize the
effect of events that are anticipated beyond their control (i.e. the effect of fuel prices
or recession) to justify their numbers for an uncertain period of time.
Other Significant Events: A firm that has undergone restructuring or report
discontinued operations may have continuing expenses for a number of years.
Also, the impact of lawsuits potentially resulting in significant settlements should be
considered. This may not necessarily be of great concern given that many
corporations face lawsuits annually.

2.4.4 Summary of Fraud Literature
To date, past works have used either financial data or qualitative data to detect fraudulent
reporting. However, considering financial data only is risky as management can make
fictitious entries to make its firm look favourable. Meanwhile, auditors who only use fraud
checklists tend to overweight cues indicative of managements characteristics when they
are not fully reliable because important factors vary by industry (Wilks et al., 2004).
25

Therefore, combining both financial and qualitative data would be a valuable contribution to
this field.

2.5 Market Beta Equity Risk and Other Factors
In addition to short-term liquidity, long-term solvency, bankruptcy and manipulation risks,
companies face risks like recessions, inflation, changes in interest rates, rising unemployment
and similar market and economic factors that affect all firms but to varying degrees,
depending on the nature of their operations (Luenberger, 1998). From the Capital Asset
Pricing Model (Figure 2.2), the reward-to-risk ratio for any security, portfolio or firm in
relation to that of the overall market is given by:
ER
i
= R
f
+
i
(ER
m
- R
f
) (2.17)
where ER
is the expected return of the firm, R

f
is the risk-free rate of interest and ER
is
the expected return of the market. ER
is often referred to as the market premium.

i
measures the sensitivity or covariability of a firms returns to the returns of diversified
portfolio of all the shares traded on the market; that is,
i
=
Cov(R
i
,R
m
)
Var(R
m
)
. It represents systematic
or nondiversifiable risk of the firm which depends on its degree of operating leverage, degree
of financial leverage, and variability of sales. Since
i
of the market portfolio is equal to 1
(Figure 2.3), a security with a
i
less than 1 is less volatile than the market, while a
i
greater
than 1 is more volatile. For example, a stock with
i
= 1.2 is theoretically 20% more volatile
than the market. In addition, a positive implies that the firms returns follow the markets
returns whereas a negative
i
implies that the firms returns move opposite to that of the
market. Finally, when
i
= 0, the firms returns change independent of the markets returns
(Table 2.10).
CAPM makes several assumptions:
Trades may occur without transaction or taxation costs.
There is perfect competition (i.e. participants cannot influence prices).
Investors are rational, risk-averse and make decisions based solely on the expected
return and variance of portfolio returns.
All information is available to all investors at the same time.
Unlimited lending and borrowing is permitted at the risk-free interest rate.
26

All assets, including human capital, are marketable and infinitely divisible.
Firm returns are normally distributed random variables and investors employ a
quadratic utility.
The variance of returns is an adequate measurement for risk.
Given a certain expected return, investors will prefer lower risk to higher risk, and
conversely, given a certain level of risk, they will prefer higher returns to lower ones.

Figure 2.2: Capital Asset Pricing Model (CAPM)

Figure 2.3: CAPM Security Market Line
Table 2.10: Definition of of Security or Company
< 0 = 0 0 < < 1 = 1 > 1
Returns move
opposite of market
returns and are less
volatile
Returns change
independent of
market returns and
are less volatile
Returns follow
market returns
and are less
volatile
Market
return
Returns follow
market returns
and are more
volatile

Capital Market Line
R
f
Expected
Return
Standard
Deviation
Efficient Frontier
M
M
Market Portfolio
Risk-Free
Asset
Risk of Market
Portfolio
E(R
M
) Expected Return on
Market Portfolio
Security Market Line
R
f
Expected
Return
M
E(R
M
)
= 1
Market Portfolio
27

Other market ratios which measure investor response to owning a companys stock and
also the cost of issuing stock include the payout ratio; P/E ratio and dividend yield (Table
2.11).
Table 2.11: Market Ratios
Payout Ratio
Dividend per Share
Earnings per Share

P/E Ratio
Market Price of Common Stock per Share
Earnings per Share

Dividend Yield
Dividend per Share
Market Price of Common Stock per Share

28

3 Background II: Profile of the Retail-Apparel Industry
3.1 Overview
Any business that sells finished merchandise to an end user is in the retail industry.
According to the U.S. Census Bureau, retail is the second largest industry (only to health
care) in the country. It can be divided into many sectors (Table 3.1) with an average gross
margin of 28%. Annual total sales were estimated at $4.45 trillion, $4.40 trillion and $4.09
trillion in 2007, 2008 and 2009, respectively accounting for 12% of the total trade volume
of all U.S. based businesses. According to the U.S. Department of Labour, the retail industry
employed nearly 15.5 million people in 2007, providing more than 11% of the total
employment in the country. However, by the end of 2008, 500,000 jobs were eliminated a
notable 30% of the 2.6 million layoffs that occurred overall. The recent decline is the effect
of the recession.
Table 3.1: Retail Industry Segmentation (US Census, 2002)

Establishments Sales (in $M)
Annual Payroll
(in $M)
Paid
Employees
Retail Trade 1,114,637 $3,056,422 $302,114 14,647,675
Motor vehicle & parts dealers 125,139 $801,7402 $64,549 1,845,496
Furniture & home furnishings
stores
65,204 $91,814 $12,843 535,029
Electronics & appliance stores 46,779 $82,228 $9,330 391,015
Building material & garden
equipment & supplies dealers
88,314 $246,561 $30,067 1,160,016
Food & beverage stores 148,804 $456,942 $48,686 2,838,653
Health & personal care stores 81,797 $177,947 $20,266 1,024,429
Gasoline stations 121,446 $249,141 $13,701 926,792
Clothing & clothing
accessories stores
149,810 $167,934 $21,391 1,426,573
Sporting goods, hobby, book,
& music stores
62,236 $73,212 $8,703 611,144
General merchandise stores 40,723 $445,225 $42,647 2,524,729
Miscellaneous store retailers 129,464 $90,812 $12,835 792,361
Non-store retailers 54,921 $172,865 $17,094 571,438

Ninety percent of retail stores are single-store businesses which make up for less than half of
total sales. The majority of revenue is generated by companies that run retail chains such as
Wal-Mart. In fact, 35 big box retailers generate approximately $335 billion each year.
These retailers also have large payrolls: Wal-Mart alone employs 1 million people in the US.
The Bureau of Economic Analysis (BEA) provides statistics on GDP by industry
(Table 3.2). Historically, the retail industrys contribution to the GDP has increased at a
compounded annual rate of 4% from 1997 to 2008. However, in 2009, the value-added fell
29

for the first time since 1991, coinciding with a 4% decrease in personal consumption
expenditures. The BEA also reports on sales and inventories (Table 3.3) which have
remained fairly constant for this industry.
Table 3.2: Value-Added by Industry, 2005-2010, in Billions of Dollars (BEA)
2005 2006 2007 2008 2009 2010
Total GDP 12,421.9 13,178.4 14,061.8 14,369.1 14,119.0 14,660.4
Retail Trade GDP 824.7 866.5 886.1 840.2 819.6 862.8

Table 3.3: Manufacturing and Trade (BEA)
2009 2010
August September October November December January
Sales ($B)
Retail Trade 316.6 309.1 312.8 318.5 317.4 316.3
Clothing and accessing stores 18.2 18.3 18.3 18.2 18.0 18.4
Inventory ($B)
Retail Trade 441.2 445.7 447.4 446.6 449.0 448.0
Inventory-Sales Ratio ($B)
Retail Trade 1.39 1.44 1.43 1.40 1.42 1.42

The segment of the retail industry of focus in this thesis is apparel. The apparel
industry supplies consumers with affordable utilitarian attire and innovative styles for those
fashion-forward. It is diverse, with hundreds of product lines varying in style, price and
demographic. The US apparel industry is large, mature, and highly fragmented, with total
sales decreasing since the onset of recession (Table 3.4). Despite over 150,000
establishments, the largest companies (which are studied in this thesis) account for 65% of
the revenue. In fact, most businesses operate one store with average annual sales of $1
million and 10 employees, and compete by targeting their specialties at different customers.
Table 3.4: Total Sales of Apparel Industry from 2001-2009 (US Department of Commerce)
2001 2002 2003 2004 2005 2006 2007 2008 2009
Sales ($B) 168 173 179 191 201 213 221 216 204

Apparel is made domestically and internationally with imports and exports shown in
Figure 3.1.
30

Figure 3.1: US Imports and Exports by Year (in Billions of Dollars) (US Department of Commerce)

3.2 SWOT Analysis
For a greater understanding of the retail-apparel industry, a survey of SWOT analysis reports
prepared by Business Source Primer on companies studied in this thesis was conducted.
SWOT analysis identifies and categorizes internal (strengths and weaknesses) and external
factors (opportunities and threats) that are favourable and unfavourable to achieving an
objective. In this context, a strength gives a firm an advantage over others (leading to
stronger financials) whereas a weakness places it at a disadvantage (leading to a decline in
revenue or profit, increase in bad debts or weakened debt service capability). An opportunity
is an external chance to make greater sales profits whereas a threat is an external factor that
could cause trouble for the business.
This section presents the strengths, weaknesses, opportunities and threats that apparel
companies generally share. Note that this list is not comprehensive and excludes financial-
related characteristics.

3.2.1 Strengths
Brand Portfolio Management and Customer Loyalty
Brands are increasingly significant to apparel as consumers are spending their disposable
income more carefully. Firms need strategies (advertising campaigns, focus groups,
licensing, loyalty programs, etc.) to create value for the consumer and brand equity for
themselves. To be competitive and to ensure that pricing is not the sole driver of purchases,
companies must incorporate tangible product features, such as quality and appearance, with
31

intangibles, such as a personal level of communication and innovation, an emotional
connection or inspirational value.

Multi-Sales Channels
Most companies distribute their products through a variety of channels: wholesale, catalogue,
and Internet sales, and their own retail stores. Of particular importance, consumers are
increasingly shopping on the Internet: U.S. e-commerce sales reached $36.6 billion or 4.1%
of total sales in 2010 (US Department of Commerce). Online sales of apparel accounted for
about $23.6 billion of sales or 17% of total US online sales in 2008 (Forrester Research,
2011). Thus, companies must establish a strong online presence through a user-friendly and
convenient website to take advantage of this trend.

3.2.2 Opportunities
Retailers can look to U.S. population projections (Table 3.5) for growth opportunities
(National Retail Federation, NRF, 2010):
Teenagers: Youth aged 15 to 19 represents 7.1% of the US population and will soon
have more income as they enter adulthood.
Generation Y: 71 million Americans born between 1977 and 1994 have or are
entering the workforce and have more to spend on apparel, especially to expand their
wardrobes to include professional attire.
Children: As Generation Y grows older, they are starting to have children, benefiting
retailers that target infants and youth.
Women: Ladies born between 1946 and 1964 make up the largest segment of the
apparel market.
Other rising trends that businesses should note include:
Apparel for sports and activities-based entertainment: People are increasingly
interested in maintaining an active and healthy lifestyle. Athletic apparel has
advanced with the creation of materials that insulate from the cold while repelling
sweat from the body, and devices that can be synched with computers to measure
performance.
32

Green initiatives and goods: Some companies have marketed themselves as socially
responsible, appealing to consumers who favour conscious spending. Despite the
faltering economy, 61% of consumers bought a green product in 2008 (NRF, 2010).
Table 3.5: US Population Projections (US Department of Commerce)
2000 2015 2030
Age Group
Number
(1000s)
% of
Total
Number
(1000s)
% of
Total
Number
(1000s)
% of
Total
Under 5yrs 21,100 6.8 22,076 6.8 23,484 6.6
5 to 14 yrs 41,281 13.3 43,365 13.3 47,225 13.2
15 to 19 yrs 21,770 7.0 21,209 6.5 23,545 6.6
20 to 24 yrs 21,779 7.0 22,342 6.9 23,168 6.5
25 to 29 yrs 21,418 6.9 22,400 6.9 22,417 6.3
30 to 34 yrs 29,400 6.9 22,099 6.8 23,699 6.6
35 to 39 yrs 20,267 6.5 20,841 6.4 23,645 6.6
40 to 44yrs 21,010 6.8 20,460 6.3 22,851 6.4
45 to 49 yrs 22,595 7.3 21,001 6.5 21,154 6.9
50 to 54 yrs 22,109 7.1 22,367 6.9 20,404 5.7
55 to 64 yrs 36,275 11.7 40,544 12.5 41,952 11.7
65 yrs and over 40,229 13.0 46,837 14.4 63,907 17.9
All ages 310,233 100.0 325,540 100.0 357,452 100.0

3.2.3 Weaknesses and Threats
Intense Competition
The apparel industry is highly competitive due to the power that consumers have to readily
switch brands and the relatively low barriers of entry. However, while it may be easy to start
a business, it is difficult to sustain as start-ups are typically undercapitalized and lack
marketing resources to build brand loyalty as well as the technological infrastructure that
major retailers now demand.

Rising Unemployment Rate, Declining Consumer Confidence and Decreasing Discretionary
Spending
A high unemployment rate, at over 10% (Bureau of Labor Statistics, 2010) because of the
recession, has decreased total disposable personal income. As a result, consumer confidence
5

(Figure 3.2) has weakened, dampening spending on apparel (Figure 3.3). In 2009, consumers
spent about $322 billion on clothing and footwear down 3.6% from 2008, when they spent
$334 billion (BEA, 2010).

5
The index represents a relative measure of how consumers feel about the strength of the economy, business trends, their
job security or employment prospects, and their future earnings prospects.
33

Figure 3.2: Consumer Confidence Index by Year (1985 = 100) (The Conference Board)

Figure 3.3: Apparels Share of Total Personal Consumption Expenditures by Year (in %) (US
Department of Commerce)

Interest Rates and Inflation
Interest rates affect consumer purchasing decisions as higher levels can curb spending as
people prioritize paying off credit card debts and reining in expenses. To a business, the cost
of borrowing influences management decisions regarding acquisitions, new product
introductions, capital expenditures, dividends and stock repurchases.
Inflation affects pricing decisions of apparel companies and their suppliers who likely
pass on cost increases to the consumer. However, while prices for many products and
services tend to rise over time, apparel prices have fallen each year from 1998 through 2005
(Figure 3.4).
34

Figure 3.4: Consumer Price Indices (1982-84 = 100) (Bureau of Labour Statistics)

Rising Costs and Reliance on Foreign Production
As raw material, freight and labour costs continue to rise, retailers outsource manufacturing
jobs to Asia, Latin America and the Caribbean, especially with increased quotas, reduced
tariffs and free-trade and preferential-trade agreements. According to the US Department of
Labour, domestic employment in apparel manufacturing peaked in 1973 at 14.5 million and
has since decreased drastically (Figure 3.5).

Figure 3.5: US Apparel Industry Employment by Year (Production workers, in millions) (Bureau of
Labour Statistics)

China accounted for 37.2% of all apparel imported into the US in 2009, rising from
32.0% in 2008 (US Department of Commerce). At the same time, more manufacturing plants
are surfacing in Mexico and the Caribbean, providing a quicker turnaround time just based
on proximity. Although foreign production drives apparel prices down for consumers, there
are risks associated with heavy reliance when the yuan appreciates and production costs
increase in China, as well as when there is political strife in Mexico.

35

3.3 Future Outlook
The recession has led to many personal and business bankruptcy filings. The Administrative
Office of the U.S. Courts compiles statistics on bankruptcy filings each year (Table 3.6). As
shown, the number of corporate filing has increased with the onset of recent recession.
Table 3.6: Bankruptcy Cases Commenced, Terminated and Pending During Twelve Month Periods
Ending December 31
Year Filings Terminations Pending
2000 1,253,444 1,238,805 1,401,552
2001 1,492,129 1,357,252 1,536,429
2002 1,577,651 1,475,810 1,637,418
2003 1,660,245 1,589,383 1,708,606
2004 1,597,462 1,661,919 1,644,149
2005 2,078,415 1,576,555 2,148,845
2006 617,660 1,435,482 1,331,023
2007 850,912 891,784 1,280,137
2008 1,117,771 1,019,426 1,378,482
2009 1,473,675 1,284,714 1,575,624
2010 1,593,081 1,512,408 1,656,340

While this Office does not provide a breakdown of bankruptcy cases by industry, the
International Council of Shopping Centres (ICSC) reports that more than 150,000 retail
stores closed in 2008, an increase from the 135,000 that went out of business in 2007.
Furthermore, bankruptcy is not limited to small stores. For a complete listing of large
companies that have filed for bankruptcy, please visit http://retailindustry.about.com/.
The apparel industry is starting to look promising again. In fact, from its low in
March 2009 to its recent high in February 2012, the S&P Apparel Retail Index has risen
136%. Despite a decline in sales, there are other positives: 1. most cost-cutting that could be
done has been done; 2. inventories are no longer declining sharply and are now aligned with
consumer demand so businesses need fewer markdowns to sell excess goods which translates
into gross margins recovering and revenue stability; and 3. retailers that closed stores and
exited marginal locations have effectively reduced fixed costs and support improved margins.
In the longer term, however, apparel sales may continue to be weak as baby boomers
prioritize funding retirement, childrens tuition and healthcare costs, over clothing. In order
to improve profitability, companies will have to operate leaner (less inventory and lower
payroll), improve customer experience and loyalty, use multichannel shopping, and/or create
private label brands (Deloitte, 2008).

36

4 Data Envelopment Analysis
In industry, measuring the performance or efficiency of a production or decision making unit
(DMU) can be a complicated proposition. Over the past thirty years, two classes of frontier
methodologies have been developed to measure DMU efficiency relative to an empirically
defined best-practice standard: the non-parametric linear programming approach and the
parametric econometric approach. These approaches differ in the assumptions made about
the shape of the efficient frontier, the treatment of random error, and the distributional
assumptions regarding inefficiency and random error. Specifically, nonparametric
approaches measure technical efficiency, focusing on levels of input relative to levels of
output. For a DMU to be technically efficient, it must either minimize its inputs given
outputs and/or maximize its outputs given inputs. On the other hand, parametric approaches
measure economic efficiency, choosing optimal levels of inputs and/or outputs based on
reactions to market prices. For a DMU to be economically efficient, its inputs and/or outputs
must be at levels that achieve cost minimization or profit maximization (Bauer et al., 1998).
Generally, the major advantages of frontier methodologies are that they can identify and
assess areas of best performance within complex operational settings, and they provide a
single aggregate measure of performance (or efficiency score) for each DMU in and relative
to the sample. Moreover, nonparametric approaches can simultaneously handle multiple
inputs and multiple outputs. However, they are often deterministic and cannot account for
any potential errors in the data, treating deviations from the efficient frontier as inefficiencies
(Cummins and Zi, 1996).
In this chapter, a thorough description of a nonparametric linear programming approach
called Data Envelopment Analysis is provided. For a brief overview of parametric
econometric approaches, please refer to Appendix A.

4.1 Introduction
Data Envelopment Analysis (DEA) is a nonparametric fractional linear programming
technique that can be used to rank and compare the relative performance of DMUs operating
under comparable conditions. It is particularly effective in handling complex processes
where DMUs use multiple inputs to produce multiple outputs. Unlike parametric
methodologies which assume that the same average equation applies to all observations or
37

DMUs, DEA optimizes each DMU, arriving at an efficiency score for every DMU relative
to the entire sample (Figure 4.1) (Cooper et al., 2007).
DMUs identified as empirically efficient define the efficient frontier: an N
dimensional surface (where N is the total number of inputs and outputs) that envelops the
inefficient DMUs in the sample. The efficient frontier provides a benchmark for the changes
in inputs and outputs necessary to render inefficient DMUs efficient; that is, inefficient
DMUs can be projected to a specific achievable target on the efficient frontier.

Figure 4.1: Data Envelopment Analysis Versus Regression Analysis
There are two main models in DEA: Charnes-Cooper-Rhodes (CCR), which is a
constant returns-to-scale model, and Banker-Charnes-Cooper (BCC), which is a variable
returns-to-scale model. Both models generate a piecewise-linear envelopment surface and
are either input- or output-oriented, depending on whether the objective is to maximize input
contraction or output expansion, with output production or input consumption, respectively,
kept constant. Both orientations yield identical envelopment (convex) surfaces but differ in
the manner in which inefficient DMUs are projected onto the efficient frontier (Figure 4.2)
(Cooper et al., 2007). Examples are provided in Appendix B.
Input
Output
*
*
*
*
*
*
*
*
DEA Efficient Frontier
Regression Line
Real Observations
(DMUs)
Best Performers
38

Figure 4.2: Inefficiencies and Projections in CCR and BCC Models with One Input and One Output

4.2 CCR Model
4.2.1 Input Orientation
The Farrell measure for efficiency of a production unit (or DMU) in the case of a single
virtual input producing a single virtual output is (Farrell, 1957):
Input Virtual
Output Virtual
=

In 1978, Charnes, Cooper, and Rhodes developed DEA as an extension of Farrells
approach to evaluating empirical efficiency for the case of multiple inputs and multiple
outputs (Figure 4.3).

Figure 4.3: DEA System with Multiple Inputs and Multiple Outputs

Known as the CCR model, a given DMU
o
has a relative efficiency
o
defined as the ratio of
the sum s of weighted outputs (y
r
for r = 1, , s) and the sum of m weighted inputs (x
i
for i =
1, , m):
Input
Output
*
*
*
*
*
*
*
BCC Frontier
*
*
Input
Inefficiency
Output
Inefficiency
Input
Output
*
*
*
*
*
*
*
CCR Frontier
*
*
Input Inefficiency
(input orientation =
minimize input
consumption, output fixed)
Output Inefficiency
(output orientation =
maximize output
production, input fixed)
y
s
y
2
y
1
x
1
x
r
x
2
DMU
2
DMU
3
DMU
1
39

mo m o o
so s o o
o
x v x v x v
y u y u y u
+ + +
+ + +
=
...
...
2 2 1 1
2 2 1 1

Input weights v
i
(i = 1, , m) and output weights u
r
(r = 1, , s) are not fixed in advance but
derived from the data such that each DMU appears as favourable as possible. An
efficiency score and weights for DMU
o
can be obtained by solving the following
optimization (Formulation 4.1). From a sample of n DMUs, each with m inputs and s
outputs, the objective of this input-oriented CCR model is to obtain the best set of weights
(v
*
, u
*
) that maximizes the ratio
*
of each DMU
o
(o = 1, , n), by minimizing observed
inputs equiproportionally while output levels remain fixed with the constraint that the
optimal efficiency for each DMU is at most 1 (
*
1). A DMU
o
is CCR-efficient only if
*
=
1 and there exists at least one optimal (v
*
, u
*
) with v
*

> 0 and u
*

> 0.
input the given to multiplier input the is
output the given to multiplier output the is
DMU input to the of amount the is
DMU from output the of amount the is Where
0 ,..., ,
0 ,..., ,
) ..., , 1 ( 1
...
...
Subject to
...
...
Maximize
th
th
th
th
2 1
2 1
1 1
1 1
1
1
2 2 1 1
2 2 1 1
1
1
i v
r u
j i x
j r y
u u u
v v v
n j
x v x v
y u y u
x v
y u
x v x v x v
y u y u y u
y u
y u
i
r
ij
rj
s
m
mj m j
sj s j
m
i
ij i
s
r
rj r
mo m o o
so s o o
m
i
io i
s
r
ro r
o
=
+ +
+ +
=
+ + +
+ + +
= =
=
=
=
=

(4.1)
The CCR model can be presented in a less computationally intensive linear
programming (LP) form: both primal and dual formulations (Formulations 4.2 and 4.3) yield
the same optimal solution. The dual formulation of the input-oriented CCR model is
expressed with a set of non-negative intensity variables
T
n
) ,..., , (
2 1
=
, representing the
weight of each of the n DMUs.

40

CCR Primal (4.2) CCR Dual (4.3)
0 ,..., ,
0 ,..., ,
) ..., , 1 ( 0
1 Subject to
Maximize
2 1
2 1
1 1
1
1
=
=

= =
=
=
s
m
m
i
ij i
s
r
rj r
m
i
io i
s
r
ro r
u u u
v v v
n j x v y u
x v
y u

0
) ..., , 1 (
) ..., , 1 ( Subject to
Minimize
1
1
=
=
=
=
j
n
j
j rj ro
n
j
j ij io
s r y y
m i x x

The optimization is performed once for each DMU in the sample with the objective of
reducing all inputs equiproportionally; that is, optimality is achieved by minimizing inputs by
a factor of . Consequently, CCR models are radial models as their goal is to adjust inputs
or outputs (in the case of output orientation) radially from the origin. However, further input
decreases or output increases may still be possible after radial optimization has been
achieved (Figure 4.4).

Figure 4.4: Input Minimization, Output Maximization and Slacks

Input excesses s
-
and output shortfalls s
+
are known as input and output slack variables,
respectively, and are optimized in a second phase:
Output
1
/Input
Output
2
/Input
Radial Output
Maximization
Output
1
Slack s
1
+
Input
1
/Output
Input
2
/Output
*
*
*
DEA Frontier
*
Radial Input
Minimization
Input
1
Slack s
1
-
*
Input
2
Slack
s
2
-
*
*
*
*
*
*
DEA Frontier
Output
2
Slack s
2
+
Input-Oriented Model Output-Oriented Model
41

) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 (
) ..., , 1 ( Subject to
Maximize
1
1
1 1
s r s
m i s
n j
s r y y s
m i x x s
s s w
r
i
j
ro j
s
r
rj r
m
i
j ij io
*
i
s
r
r
m
i
i
=
=
=
= =
= =
+ =
+
=
+
=
=
+
=

(4.4)
where
*
is the optimal radial contraction computed from the first phase (Formulation 4.3).
Thus, a DMU is defined as fully efficient if and only if
*
= 1 (radial or technical
efficiency) and s
+*
= s
-*
= 0 (zero slacks). It is considered Pareto-Koopman efficient if and
only if it is not possible to improve any input or output without worsening some other input
or output (Charnes et al., 1978). If only
*
= 1 is satisfied, then the DMU is weakly
efficient, with the mix inefficiencies associated with nonzero slacks. Note that fully and
weakly efficient units are radially efficient regardless of slacks.
An inefficient DMU can be improved by referring its inefficient behaviour to the
efficient frontier formed by E
o
, the reference set of DMU
o
composed of efficient DMUs.
This improvement is a projection to the point
on the frontier where:

) ..., , 1 (
) ..., , 1 (
*
s r y y s y y
m i x x s x x
ro
E j
*
j rj r ro ro
io
E j
*
j ij
-*
i io
*
io
o
o
= = + =
= = =

(4.5)
are the coordinates of a virtual linear composite DMU (i.e. DMU

i
i
where DMU
i
s
are efficient and
i
are proportionality weights for DMU
i
) used to evaluate the performance of
DMU
o
. It represents the target for efficient production that DMU
o
strives for (Cooper et al.,
2005).

4.2.2 Output Orientation
The output-oriented CCR model aims to maximize outputs without any addition to the
observed input values. The primal and dual formulations are:

42

CCR Primal (4.6) CCR Dual (4.7)
0 ,..., ,
0 ,..., ,
) ..., , 1 ( 0
1 Subject to
Minimize
2 1
2 1
1 1
1
1
=
=

= =
=
=
s
m
m
i
ij i
s
r
rj r
s
r
ro r
m
r
io i
q q q
p p p
n j x p y q
y q
x p

0
) ..., , 1 (
) ..., , 1 ( Subject to
Maximize
1
1
=
=
=
=
j
n
j
j rj ro
n
j
j ij io
s r y y
m i x x

In relation to the input-oriented model, p
*
=
v
*
*
, q
*
=
u
*
*
,
*
=
*
and
*
=
1
*
. Consequently,
*
1
(i.e. the inverse of
*
1). The input t
-
and output t
+
slacks (Equation 4.8) of the output-
oriented model are also introduced in a second phase (Formulation 4.4):
) ..., , 1 (
) ..., , 1 (
*
1
1
s r y y t
m i x x t
ro j
s
r
rj r
m
i
j ij io i
= =
= =

=
+
=

(4.8)
where
*
is the optimal expansion from the first phase (Formulation 4.7) and t
-*
=
s
-*
*
and
t
+*
=
s
+*
*
.
A DMU is fully efficient if and only if
*
= 1 and all optimal slacks are zero. For
inefficient DMUs, the CCR projection
is:
) ..., , 1 (
) ..., , 1 (
* *
s r t y y
m i t x x
r ro ro
-*
i io io
= + =
= =
+

(4.9)
Finally, the CCR model assumes that the production process is governed by constant
returns-to-scale (CRS) which states that if (x
o
, y
o
) is a possible combination of input and
output activity, then (kx
o
, ky
o
) is also feasible for every positive scalar k (Charnes et al.,
1978). A CRS efficient frontier is one hyperplane, such that for every increase in inputs,
there is a proportionate increase in outputs.

4.3 BCC Model
Banker, Charnes and Cooper modified the CCR model for variable returns-to-scale (VRS).
The frontier of this BCC model is comprised of piecewise linear and concave segments
which exhibit increasing returns-to-scale, followed by constant returns-to-scale, then
43

decreasing returns-to-scale (Figure 4.5). The BCC model also has two skirts (shown as
dotted lines) that run parallel to the input and output axes. These are virtual segments that
allow for calculations of input excesses s
-
and output shortfalls s
+
. Like the CCR model, the
BCC model is radial and either input- or output-oriented (Banker et al., 1984).

Figure 4.5: Constant versus Variable Returns-to-Scale (RTS)

4.3.1 Input Orientation
The primal and dual formulations for the BCC input-oriented model are:
BCC Primal (4.10) BCC Dual (4.11)
sign in free
0 ,..., ,
0 ,..., ,
) ..., , 1 ( 0
1 Subject to
Maximize
2 1
2 1
1 1
1
1
o
s
m
o
m
i
ij i
s
r
rj r
m
i
io i
o
s
r
rj r
u
u u u
v v v
n j u x v y u
x v
u y u z
=
=
=

= =
=
=

0
1
) ..., , 1 (
) ..., , 1 ( Subject to
Minimize
1
1
1
=
=
=
=
=
=
j
n
j
j
n
j
j rj ro
n
j
j ij io BCC
BCC
s r y y
m i x x
BCC
is the proportional reduction that can be applied to all inputs to improve BCC
efficiency. As previously, slacks (Equation 4.12) can be incorporated in a second phase
(Formulation 4.4):
) ..., , 1 (
) ..., , 1 (
1
1
*
s r y y s
m i x x s
ro j
s
r
rj r
m
i
j ij io BCC i
= =
= =

=
+
=

(4.12)

A DMU is BCC-efficient if and only if
*
BCC
= 1 and has zero slacks (s
+*
= s
-*
= 0).
CCR efficiency can never exceed BCC efficiency; a DMU found to be efficient in CRS will
Input
Output
*
*
*
*
*
*
*
*
BCC Frontier
CCR Frontier
Increasing
RTS
Decreasing
RTS
Constant
RTS
44

also be efficient in VRS, but the converse is not necessary true. The BCC input-oriented
projection

is:
) ..., , 1 (
) ..., , 1 (
*
*
s r s y y
m i s x x
r ro ro
-*
i io BCC io
= + =
= =
+

(4.13)
As can be seen, the BCC model is identical to the CCR model except for the addition
of the convexity constraint,
j
n
j=1
=1 or the variable u
o
, which is the unrestricted dual
variable in the primal formulation. This constraint reduces the feasible region for the LP
from a convex cone defined by the DMUs to the convex hull covering all the DMUs, thereby
increasing the number of efficient DMUs (Charnes et al., 1994).

4.3.2 Output Orientation
The primal and dual formulations of the output-oriented BCC model are presented in
Formulations 4.14 and 4.15, respectively (Charnes et al., 1994).

BCC
is the proportional
augmentation in all of the outputs that represents technical, radial efficiency, while v
o
is the
dual variable associated with the convexity constraint in the primal problem.
BCC Primal (4.14) BCC Dual (4.15)
sign in free
0 ,..., ,
0 ,..., ,
) ..., , 1 ( 0
1 Subject to
Minimize
2 1
2 1
1 1
1
1
o
s
m
o
s
r
rj r
m
i
ij i
s
r
ro r
m
r
o io i
v
q q q
p p p
n j v y u x v
y u
v x v z
=
=
=

= =
=
=

0
1
) ..., , 1 (
) ..., , 1 ( Subject to
Maximize
1
1
1
=
=
=
=
=
=
j
n
j
j
n
j
j rj ro BCC
n
j
j ij io
BCC
s r y y
m i x x

Slacks (Equation 4.16) are accounted for in a second phase (Formulation 4.4) after
maximal augmentation
BCC

(Formulation 4.15).
) ..., , 1 (
) ..., , 1 (
*
1
1
s r y y t
m i x x t
ro BCC j
s
r
rj r
m
i
j ij io i
= =
= =

=
+
=

(4.16)
A DMU is BCC-efficient if and only if
*
BCC

and t
+*
= t
-*
= 0, while an inefficient
DMU can be improved with the projection:
45

) ..., , 1 (
) ..., , 1 (
* *
s r t y y
m i t x x
r ro BCC ro
-*
i io io
= + =
= =
+

(4.17)

4.4 Additive Model
Both the CCR and BCC models require a distinction between input and output orientation.
However, the additive (ADD) model combines both orientations (Figure 4.6) with
Formulations 4.18 and 4.19.

Figure 4.6: Additive Model
ADD Primal (4.18) ADD Dual (4.19)
sign in free
1 ,..., ,
1 ,..., ,
) ..., , 1 ( 0 Subject to
Maximize
2 1
2 1
1 1
1 1
o
s
m
o
s
r
ij j
m
i
ij i
o
s
r
jo j
m
i
io i
u
u u u
v v v
n j u y u x v
u y u x v w
= +
+ =

= =
= =

) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 ( 0
1
) ..., , 1 (
) ..., , 1 ( Subject to
Maximize
1
1
1
1 1
s r s
m i s
n j
s r y s y
m i x s x
s s z
r
i
j
n
j
j
ro r j
n
j
rj
io i
n
j
j ij
s
r
r
m
i
i
=
=
=
=
= =
= = +
+ =
+
=
+
=
=
=
+
=

z and w reflect all inefficiencies in both inputs and outputs, with the efficiency score
measured implicitly in the slacks. A DMU is ADD-efficient if and only if the DMU has zero
slacks (s
+*
= s
-*
= 0); there are no restrictions for z and w. Inefficient DMUs can be improved
by ADD projection
:
Input
Output
*
*
*
*
DEA Efficient Frontier
Input
Orientation
Output
Orientation
ADD
Projection
*
s
+
s
-
46

) ..., , 1 (
) ..., , 1 (
*
s r s y y
m i s x x
r ro ro
-*
i io io
= + =
= =
+

(4.20)
The ADD model is particularly useful for applications that have negative data in the
inputs and outputs because it is translation invariant: a DEA model is translation invariant if
translating the original input and/or output values results in a new problem that has the same
optimal solution as the original (Figure 4.7). In other words, efficiency evaluation is
independent of the origin of the coordinate system, and with respect to the unit of
measurement of each input and output. ADD is translation invariant in both inputs and
outputs when the convexity constraint is added, as opposed to input-oriented models which
are translation invariant with respect to outputs only, and output-oriented models which are
invariant under the translation of inputs.

Figure 4.7: Translation Invariance in the Additive Model

4.5 Slacks-Based Measure of Efficiency Model
An extension of the additive model is the slacks-based measure of efficiency (SBM) model
(Formulation 4.21). The main difference between the ADD and SBM models is that the
former is an absolute measure (a summation of slacks) while the latter computes an
efficiency score ] 1 , 0 [ as a ratio of average relative input consumption to average relative
Input
Output
*
*
*
*
*
x
y
O
*
*
*
*
*
s
+
s
-
s
+
s
-
s
+
Original
Frontier
Translated
Frontier
47

output production. Consequently, SBM models are units invariant
6
(unlike ADD models),
but not translation invariant (Tone, 2001).
1 0 Where
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 (
) ..., , 1 ( Subject to
/
1
1
/
1
1
Minimize
1
1
1
1

=
=
=
= =
= = +
+
=
+
+
=
=
=
+
=
s r s
m i s
n j
s r y s y
m i x s x
y s
s
x s
m
r
i
j
ro r j
n
j
rj
io i
n
j
j ij
s
i
ro r
m
i
io i

(4.21)
A DMU is SBM-efficient when
*
= 1 (zero slacks). Inefficient DMUs can be improved with
the same projection as ADD (Equation 4.20). To solve the SBM model, a positive scalar
variable t must be introduced (Formulation 4.22) then linearized (Formulation 4.23) such that
- -
s S t = ,
+ +
= s S t and t = :
0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 (
) ..., , 1 (
/
1
1 Subject to
/
1
Minimize
1
1
1
1
>
=
=
=
= =
= = +
+ =
=
+
+
=
=
=
+
=
t
s r s
m i s
n j
s r y s y
m i x s x
y ts
s
t
x ts
m
t
r
i
j
ro r j
n
j
rj
io i
n
j
j ij
s
i
ro r
m
i
io i

(4.22)
0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 ( 0
) ..., , 1 (
) ..., , 1 (
/
1
1 Subject to
/
1
Minimize
1
1
1
1
>
=
=
=
= =
= = +
+ =
=
+
+
=
=
=
+
=
t
s r S
m i S
n j
s r ty s y
m i tx s x
y S
s
t
x S
m
t
r
i
j
ro r j
n
j
rj
io i
n
j
j ij
s
i
ro r
m
i
io i

(4.23)

The optimal solution is defined by
*
=
*
,
*
/ t
* *
= ,
*
/ t
* - * -
S s = ,
*
/ t
* *
S s
+ +
= .

6
A DEA model is considered units invariant if the efficiency scores are independent of the units in which the inputs and
outputs are measured provided these units are the same for every DMU.
48

4.6 Sensitivity Analysis
There are three approaches to sensitivity analysis in DEA.
1. Varying the values of inputs and outputs: test the robustness of the model by changing
the extreme values (minimums and maximums) of inputs and outputs.
2. Adding and removing variables: include or exclude an input or output variable, one at a
time, to observe the effect on the scores.
3. Searching for outliers and super-efficiencies: remove DMUs that are highly referenced
as they may be super-efficient outliers that skew the frontier.

4.7 Summary of the Basic DEA Models
Table 4.1 summarizes some important topics to consider when choosing between basic DEA
models (Cooper et al, 2007). In this table, Semi-Pos is notation for semi-positive or
nonnegative with at least one positive element in the data for each DMU, whereas Free
allows negative, zero or positive data.
*
is the efficiency score of input-oriented models
while its reciprocal (
*
1) is for output-oriented models. Technical and Mix Efficiency
indicates whether a model measures technical efficiency or mix efficiency. CRS and
VRS denote constant and variable returns to scale, respectively; for ADD and SBM
models, this depends on the addition of the convexity constraint (i.e.
i
=1).
Table 4.1: Summary of Model Characteristics (Cooper et al., 2007)
Model CCR BCC ADD SBM
Orientation Input Output Input Output None None
Data X Semi-Pos Semi-Pos Semi-Pos Free Free Semi-Pos
Y Free Free Free Semi-Pos Free Free
Translation
Invariance
X No No No Yes Yes
a
No
Y No No Yes No Yes
a
No
Units Invariance Yes Yes Yes Yes No Yes
*
[0, 1] [0, 1] (0, 1] (0, 1] None [0, 1]
Technical or Mix
Efficiency
Technical Technical Technical Technical Mix Mix
Returns to Scale CRS CRS VRS VRS C(V)RS
b
C(V)RS
b
a
The Additive model is translation invariant only when the convexity constraint is added.
b
C(V)RS means Constant or Variable returns to scale according to whether or not the convexity constraint is included.

4.8 DEA Characteristics
4.8.1 Strengths
DEA has several strengths over other frontier methodologies and techniques used in
performance measurement, such as regression analysis. First, regression analysis describes
49

the average behaviour of all units (DMUs) or maximal variance within the dataset. DEA
identifies best practice units (based on inputs consumed/outputs produced) as benchmarks
and potential improvements that are real and feasible for inefficient units to target.
Furthermore, because DEA measures empirical efficiency based on observations, no
assumptions are made as to what productions are theoretically possible (Figure 4.8).
However, in the past decade, researchers have developed approaches, such as Hall and Simar
(2002) who proposed a technique for estimating boundary points in the presence of noise, to
predict the form of a theoretical frontier.

Figure 4.8: True (Theoretical) Frontier versus DEA (Empirical) Frontier
Second, DEA is nonparametric. Unlike regression, it can simultaneously handle
multiple inputs and outputs which can be non-discretionary or exogenous (those over which
DMUs have no control) (Charnes et al., 1994). Also, it does not require the specification of
weights a priori, but calculates the optimal weightings for each DMU in order to maximize
its efficiency score, provided that all other DMUs satisfy feasibility (
*
1,
*
1).
Third, DEA can handle different types of units for the variables in the same model.
In fact, because the first phase of both the BCC and CCR models are units invariant,
variables with different units can be considered at the same time. The second stage,
however, is not units invariant but can be rendered so by modifying slacks in the objective
function by weights 1/w
i
+
or 1/w
i
-
, where w
i
+(-)
is the amount of output
i
(input
i
) for DMU
o
, or
the inter-percentile range, average, or standard deviation of the amount of output
i
(input
i
)
employed by all DMUs.
Finally, DEA produces a single all-encompassing efficiency score that characterizes a
units production of all relevant outputs. This facilitates the ranking and comparison of
Input
Output
*
*
*
*
*
*
*
*
Empirical Frontier
Theoretical Frontier
50

DMUs in multiple output frameworks, as opposed to having a separate ratio measure for each
output.

4.8.2 Limitations
First, standard DEA models do not account for random error; the assumption is that
deviations from the frontier are due to inefficiency only and makes no allowances for
measurement error or other noise. The fact that random errors may alter the shape of the
frontier, resulting in inaccurate scores, has led to the development of several stochastic DEA
models. For example, Land, Lovell and Thore (1993) accommodated uncertainty by
introducing a set of chance constraints stating that observed outputs must not exceed best
practice outputs more than 5% of the time (when inputs are predetermined). However, their
model also requires information on the joint probability distribution of variables.
Second, DEA is unable to accurately model small sample sizes because analyses
containing less than the recommended number of DMUs yield higher efficiency scores with
many DMUs appearing on the frontier. The rough rule of thumb is n maxm s, 3(m + s)
where n, m and s are the number of DMUs, inputs and outputs, respectively (Banker et al.,
1986).
Third, DEA provides only a relative efficiency score based on the set of DMUs
studied. If certain key DMUs (which are highly efficient) are excluded, then the scores
provided will not be as accurate. Because an empirical efficient frontier represents the best
practices observed within the sample, DEA provides little insight into the location of the
theoretical frontier (best performance actually possible). Thus, standard DEA models do not
identify possible improvements and targets for those DMUs found to be efficient.
Fourth, standard DEA models have complete freedom to assign the relative weights for
inputs and outputs. Consequently, DEA can assign zero or near-zero weights to some inputs
and outputs in order to maximize the efficiency of individual DMUs, especially to outliers
that consume a very small amount of one particular input, or those that produce a very large
amount of one particular output. Realistically, however, it is possible to impose constraints
on the relative magnitude of the weights via the Assurance Region method or cone ratio
methods. As an example, the ratio of weights for Input 1 and Input 2 may be restricted as
L
1,2

v
2
v
1
U
1,2
where L
1,2
and U
1,2
are lower and upper bounds. Generally, Assurance
51

Region efficiency scores are worsened by the addition of constraints; a DMU otherwise
characterized as efficient may be found to be inefficient (Thompson et al., 1986).
Weight restrictions have also been used to capture the time lag between the
consumption of inputs and the production of outputs (zpeynirci and Kksalan, 2007).
Standard DEA may sometimes be incorrect in scenarios where producing outputs in a given
period is possible by consuming inputs during several time periods. Moreover, this
production lead time may be unknown, deterministic or controllable. This led to the
introduction of dynamic DEA where Sengupta (1996) measured the efficiency of systems
by analyzing two aspects of changes over time: incremental capital inputs or investments
associated new technology, and adjustments costs, when firms tend to adapt their short-run
behaviour to the long-run optimal goals. Dynamic or time-varying efficiency scores may
assist in planning and decision-making as to when to start production, how long to produce,
which storable inputs to store, and which intermediate outputs to produce them.
Similarly, network DEA was introduced to combat the black box effect of DEA;
that is, inputs enter and outputs exit a DMU with no consideration of the intervening
steps that may be sources of inefficiency within it. Lewis and Sextons model (2004)
handled DMUs that consisted of a network of sub-DMUs, some consuming resources
produced by other sub-DMUs and some producing resources consumed by other sub-DMUs.
Perhaps of greatest significance, the network DEA Model allowed individual DMU managers
to focus efficiency-enhancing strategies on the individual stages of the production process
(Lewis and Sexton, 2004).
Finally, the use of DEA is problematic when one or more of the input or output
variables are in a ratio form (Hollingsworth et al., 2003). For example, consider a system
where single input x produces two outputs y
1
and y
2
and 4 DMUs form the efficient frontier:
A, B, C and D (Figure 4.9). DMU E is inefficient and according to the convexity axiom of
DEA, the convex combination of A and E, a DMU AE, should lie in the production
possibility set. However, if either output is in ratio form, the actual convex combination of
the A and E should be calculated as a weighted combination of two DMUs, which may fall
outside the feasible region (as illustrated by DMU AE'). Therefore, the convexity
assumption may fail, leading to incorrect efficiency scores and distorted projections
52

(Emrouznejad and Amin, 2007). The best way to rectify this problem is to use absolute
values if both the numerator and the dominator of the ratio variables are known.

Figure 4.9: One Input, Two Output Example for Use of Ratios in DEA

4.9 DEA Studies in Bankruptcy
The advantages of DEA and the intuitive relationship between inefficiency and failure have
inspired many studies on DEA and failure, mainly with banks only a few of which will be
discussed here
7
.
Barr et al. (1993, 1994) incorporated the international bank-rating system CAMELS
(Capital adequacy, Asset quality, Management quality, Earnings, Liquidity, Sensitivity to
market risk) with DEA to predict bank failure, relating it to inefficiency. The analysis of
these 6 factors for 930 banks over 5 years validated the use of DEA as surviving institutions
scored significantly different than those that failed. However, the results were specific only
to banks and not general corporations.
In another study, Kao et al. (2004) applied DEA to predict bankruptcy of 24
Taiwanese commercial banks based on 3 inputs (total deposits, interest expenses, non-
interest expenses) and 3 outputs (total loans, interest income, non-interest income). They
administered a questionnaire to general managers asking for their forecasts of financial data
for the following year. A DEA model using interval data to account for pessimistic and
optimistic views, generated efficiency scores for each bank based on these managerial
predictions. These scores were then compared to the true efficiency scores calculated from
data available in financial statements a year later, and were within the ranges of predicted

7
Please refer to the paper by Ravi Kumar et al. (2006) for a comprehensive survey of past studies.

0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
D
C
B
A
E
AE'
AE
Feasible
Region
%y
1
y
2
2
1
2
1
) (
) 1 (
= + =
+ =
E A
E A AE
if
) (
) (
%
, 1 , 1 2
1
, 1 , 1 2
1
, 1
, 1
, 1
E A
E A
AE
AE
AE
Z Z
Y Y
Z
Y
y
+
+
= =
53

scores from financial forecasts. Nonetheless, despite the value of human judgment and
hypothesis, this type of prediction was still subjective.
Extending beyond banks to general companies, Punik et al. (2008) investigated
whether technical and cost efficiency as measured by DEA can predict small business
failures in Slovenia. However, little information was provided on the methodology,
particularly with respect to how data for all years and companies from different industries
were accounted for. Also, the choice, definition and orientation of variables were not explicit
beyond the use of financial ratios a practical problem as companies may perform
differently and a mathematical limitation of DEA. Consequently, results seemed
counterintuitive: there was no difference in the rate of return on sales between surviving and
failed firms; the average value of the rate of return on equity increased for bankrupt firms;
and the absolute values of liquidity ratios were not critical even for failed firms, suggesting
that a firm with high liquidity could go bankrupt. Similarly, Cielen et al. (2004) applied 11
financial ratios to DEA to predict bankruptcy between 1994 and 1996 with Type I and II
errors of 24% and 21%, respectively. However, the number of companies, and the industries
and countries under study were not mentioned only that 276 non-bankrupt and 90 bankrupt
annual reports were examined. Also, Premachandra et al. (2011) investigated over 1000 US
companies from a spectrum of industries between 1991 and 2004. Seven ratios were applied
to an additive model with the generated DEA scores classified as bankrupt or non-bankrupt
based on a single cut-off value. This model was relatively weak in predicting corporate
failures compared to healthy firms, highlighting a common problem among bankruptcy
prediction using cut-off values: the trade-off between Type I and II errors.
DEA has also been used in conjunction with other techniques to predict bankruptcy.
For instance, Xu et al. (2009) studied corporations listed on the Shanghai Stock Exchange
and proposed a financial failure prediction model using efficiency as a predictor variable.
Specifically, the score generated by DEA (with total assets, total liabilities and cost of sales
as inputs, and income of sales as the output) was a proxy for the relative efficiency of
business operations. It was combined with 7 ratios and then applied to support vector
machines, logistic regression, multiple discriminant analysis. Results showed that using
DEA efficiency as a predictor improves prediction accuracy. However, these were limited
by an unrepresentative sample of firms: half of the 120 firms had filed for bankruptcy
54

between 1999 and 2005. Furthermore, the selection of the 7 ratios as variables was based on
the results of the t-test only, which assumed that the ratios were normally distributed.
While literature in bankruptcy has been enriched by work with DEA, prediction has
also offered opportunities to develop new DEA models. For example, in a study of
manufacturing companies, Paradi et al. (2004) analyzed a series of financial variables with
their novel worst practice DEA model
8
. A small and equal number of good and bad
performing firms showed very encouraging results. However, after more extensive analysis
with a larger dataset to reflect real proportions, the model did not work as well. Following a
similar philosophy, Shetty et al. (2012) assessed bankruptcy of IT companies operating in
India by creating an inefficient frontier based on 10 financial ratios. Although the authors
stated that their bankruptcy scores were consistent with poor performing firms, there was no
discussion of the reliability or validity of their model (i.e. no reports of misclassification or
success rates).
And in a final example, Sueyoshi et al. (1998, 2009) developed and applied DEA-DA
to assess bankruptcy. DEA-DA was derived from the DEA-Additive model and incorporated
the unique features of DEA into the framework of discriminant analysis (DA). It was
designed to find an overlap between two groups in a first stage, then in a second stage,
determine a piecewise linear classification function to further separate the groups. Sueyoshi
et al. (2009) compared DEA and DEA-DA with a dataset of 609 firms (951 non-bankrupt and
50 bankrupt). Although the input and output variables were ratios, they made a valuable
finding: DEA models misclassified all bankrupt firms because of the data imbalance;
however, DEA-DA can potentially deal with this imbalance by controlling the importance"
of the two groups, as determined by size.

8
Where normal DEA selects potentially distressed firms by measuring how inefficient they are at being good, worst practice
DEA picks out distressed firms based on how efficient they are at being bad. This approach is particularly fitting for credit
risk evaluation where the worst companies need to be clearly identified (Section 8.1.1).
55

5 Objectives
The objective of this research is to develop a Data Envelopment Analysis (DEA)-based
model which predicts the likelihood of failure of American retail-apparel companies and
suggests preventative measures based on data available from financial statements and their
accompanying Notes (which provide hints on managerial decision-making) as well as market
and economic influences. It is hypothesized that the inclusion of variables that reflect
managerial decision-making, and market and economic factors, enhance the predictive power
of mathematical models that consider financial data exclusively. For example, the industry-
standard Altman model has an accuracy of only 70%, 58% and 50% when predicting cases
today, from one year back and two years back, respectively. This work is an opportunity for
cross-disciplinary contributions to the fields of finance, accounting, and operations research
as novel metrics are created and the shortcomings of current approaches are addressed
(Figure 5.1). Also, although analysis in this thesis is specific to retail-apparel, the
development of a methodology to predict bankruptcy which can then be adapted for another
industry is of great value.

Figure 5.1: Current Limitations
- ratios are simplistic
- lack effects of
economy, market,
fraud, and managerial
decision-making
-do not provide
targets for bankruptcy
prevention
- inherently constant
returns-to-scale
- sometimes do not
have a sample with
real life proportion of
active and bankrupt
firms in order to
improve prediction
- difficulty handling
inputs and output in
ratio form
- prediction is a
work in progress
Limitations of
Data Envelopment Analysis
(DEA)
Limitations of Current
Indicators of
Bankruptcy Prediction
Limitations of
DEA Bankruptcy Studies
- focus mainly on banks
- data from financial statements
only and are ratios
- managerial decision-making
excluded
- lack market, economic and
fraud risk
- methodologies presented
unclearly
- counterintuitive results
- possible improvements not
exploited
- sometimes do not have a
sample with real life proportion
of active and bankrupt firms
56

6 Data Collection, Preparation and Exploration
6.1 List of Retail-Apparel Companies
The US retail-apparel industry
9
was selected because 1. it contains a large sample of public
companies of roughly the same size or market capitalization operating in the same country; 2.
there is competition such that turnover of companies is not uncommon; 3. while firms within
it may not sell the same products, they employ similar business models; and 4. the industrys
working environment has remained relatively consistent over the time of study to avoid
modeling the effect of external factors (i.e. the significant influence of advancing technology)
other than inflation or recession.
Originally, 85 companies in the retail-apparel industry that are actively or have been
publicly traded at some time since 1988 were considered (Table 6.1).
Table 6.1: List of Companies by Ticker Code in Retail-Apparel Industry
Company State Company State Company State
AENA Active DSW Active LOEHQ Bankrupt (1999)
AEO Active DUSS Active LTD Active
ANF Active EBHI Bankrupt (2009) LULU Active
ANN Active EDBRW Bankrupt (1996) MGR Bankrupt (1994)
ARO Active EXLA Active MW Active
BEBE Active FEETQ Bankrupt (2000) NWY Active
BKE Active FINL Active ONPRQ Bankrupt (2004)
BKRS Active FL Active PHRS Bankrupt (2000)
BSMT Bankrupt (1999) FOH Bankrupt (2003) PLCE Active
BSST Active (1997) FTAR Bankrupt (2002) PRVU Active
CACH Active FTUS Bankrupt (2004) PSS Active
CATH Active (2000) GADZQ Bankrupt (2005) PSUN Active
CBK Bankrupt (1996) GCO Active ROST Active
CHIC Active GDYS Active (2005) SCVL Active
CHRS Active GPS Active SHOE Bankrupt (2008)
CHS Active GYMB Active SJKI Active
CMALQ Bankrupt (2001) HOTT Active SKFB Active
CML Bankrupt (1998) HRLS Bankrupt (2008) SMRT Active
CMRG Active IBI Active (2002) SSI Active
CSS Bankrupt (1997) JAYJ Bankrupt (1994) SYMS Active
CTMEQ Bankrupt (1996) JCG Active TJX Active
CTR Active JOSB Active TLB Active
CTRN Active JWN Active TMAN Bankrupt (1996, 2002)
CWTR Active KDRH Bankrupt (1993) URBN Active
DABR Active (2000) KGM Active (1999) URGI Active (2007)
DBRN Active KNWN Bankrupt (1994) WALK Active
DEBS Active (2007) KUHM Active WTSL Active
DEST Active LAM Bankrupt (1995) ZUMZ Active
DLIA Active

9
SIC Division G (Retail Trade), Major Group 56 (Apparel and Accessory Stores)
57

The 61 firms that never declared bankruptcy between 1988 and 2009 were categorized as
active while the 24 firms that declared bankruptcy at least once (in the year(s) in
parentheses) were categorized as bankrupt. Common reasons for filing Chapter 11
included a heavy debt burden, change in management, fraud and recession (Table 6.2,
Appendix C). Firms that merged with or were acquired by another (in italics in Table 6.1)
were re-classified as either active or bankrupt depending on their state at the year of
transition (in parentheses).
Table 6.2: Reasons for Chapter 11 Filings Cited in Press Releases
Reason Companies Citing Given Reason
Illiquidity and/or insolvency 16
Unprofitability, net/operating losses, negative cash flows 14
Competition 9
Declining sales 8
Recession 8
Vendors refusing to deliver merchandise resulting in low inventory 5
Management changes 3
Excess inventory 2
Accounting fraud or discrepancies 2

With a list of companies, the search for data followed through federal agencies and the
University of Toronto Libraries. EDGAR is the most popular database of online corporate
information. The Electronic Data-Gathering, Analysis and Retrieval system performs
automated collection, validation, indexing, acceptance and forwarding submissions by
companies who are required by law to file forms with SEC. While it is free for public use,
EDGAR has several limitations. First, not all SEC filings before May 6, 1996 are available
although as of this date, all public companies must file on EDGAR. Second, actual annual
reports to shareholders need not be submitted on EDGAR. However, Form 10-K or 10-KSB,
which contains much of the same information as the annual report, is required. Third, the
presentation of the filings (often .pdf) is not in a manageable format (such as in Microsoft
Excel spreadsheets) ideal for mathematical analysis.

6.2 Definition of Study Period, Comparable Units and Company State
A challenge in bankruptcy research is that the data from companies do not all overlap
entirely over the same time period; that is, the start and end dates vary by company
(Figure 6.1 and Figure 6.2). All studies circumvent this by working with a smaller time
frame where there are data for each company for each year but this reduces the sample size
58

considerably. Because creating a technique that calibrates for time is beyond the scope of
this thesis, to consistently compare companies in the same environment, the period between
1996 and 2009 was studied. At best, the effect of inflation was considered but appeared to be
insignificant to the results (Section 10.7)

Figure 6.1: Time Frame of Available Data for Companies AENA-JWN
Highlighted Box Colour: Green = Active; Pink = 3 years prior/post bankruptcy
Bolded Font Colour: Red = Year of bankruptcy; Blue = Year merged/acquired
Numbers: 1 = Managerial decision-making data available; 0 = Managerial decision-
making data unavailable
Letters: A = Year merged/acquired, now active without managerial decision-making data;
B = Year of bankruptcy without managerial decision-making data;

Company 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
AENA 0 0 1 0
AEO 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
ANF 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
ANN 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
ARO 0 0 0 1 1 1 1 1 1 1 1
BEBE 0 0 0 1 1 1 1 1 1 1 1 1 1 1
BKE 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
BKRS 0 0 0 0 1 1 1 1 1
BSMT 0 0 1 1 1 0
BSST 0 0 0 1 A
CACH 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
CATH 0 0 0 1 1 1 A
CBK 0 0 0 B 1 1 1 1 1 1 1 1 1 1 1 1 1
CHIC 0 0 0 1 1 1 1 1 1 1 1 1 1
CHRS 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
CHS 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
CMAL Q 0 0 1 1 1 1 1 1 1 1 0
CML 0 0 1 1 1 1 1
CMRG 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
CSS 0 0 B 0
CTME Q 0 0 1 1
CTR 1 1 1 1 1 1 1
CTRN 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 1 1 1 1
CWTR 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1
DABR 0 0 0 1
DBRN 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
DEBS 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
DEST 1 1 1 1 1 1
DLIA 0 0 1 1 1 1
DSW 0 0 0 0 1 1 1 1
DUSS 0 1 1 1
EBHI 0 0 1 1 0 1
EDBR W 0 0 1 1 1 1 1
EXLA 1 1 1
FEET Q 0 0 1 1 1 B
FINL 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
FL 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
FOH 0 0 1 1 1 1
FTAR 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
FTUS 0 0 0 1 1 1 1 1 1 1 1 1
GADZ Q 0 0 1 1 1 1 1 1 1 1 B
GCO 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
GDYS 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 B
GPS 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
GYMB 1 1 1 1 1 1
HOTT 1 1 1 1 1 1
HRLS 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
IBI 1 1 1 1 1 1 1 A
JAYJ B 0 0 0 1 0
JCG 1 1 1 1 1 1
JOSB 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
JWN 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
59

Figure 6.2: Time Frame of Available Data for Companies KDRH-ZUMZ
Highlighted Box Colour: Green = Active; Pink = 3 years prior/post bankruptcy
Bolded Font Colour: Red = Year of bankruptcy; Blue = Year merged/acquired
Numbers: 1 = Managerial decision-making data available; 0 = Managerial decision-
making data unavailable
Letters: A = Year merged/acquired, now active without managerial decision-making data;
B = Year of bankruptcy without managerial decision-making data;

Here, a DMU (Decision Making Unit or Unit of Comparison) was associated with both a
company and a year. Each DMU had a corresponding state (Table 6.3). The reason for
predicting up to 3 years back was to be consistent with other past studies. In some analyses, a
DMU was assigned a less specific state of bankruptcy: if it was 3 years or less from
bankruptcy, then it was simply classified as bankrupt (i.e. instead of B-x where x = 1, 2 or
3, it was just B).
Table 6.3: Possible DMU States
Notation Definition Colour in Figure 6.1 and Figure 6.2
A Active, non-bankrupt year Green
B-3 Three years prior to bankruptcy Pink
B-2 Two years prior to bankruptcy Pink
B-1 One year prior to bankruptcy Pink
B Bankruptcy Year Pink

Company 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
KDRH B 0 0 1 0 1 0
KGM 0 0 1 1 1
KNWN B 1 0 0
KUHM 1 0
LAM 0 0 0 1 1 1 1 1
LOEH Q 1 1 1 1 1 1
LTD 1 1 1 1 1 1
LULU 0 0 0 1 1
MGR 0 0 1 1
MW 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
NWY 0 1 1 1 1 1
ONPR Q 0 0 1 1 1 1 1 1 1 1 1 B
PHRS B 0 0 1 1 1 1 B
PLCE 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1
PRVU 0 0 0 1 1 1 1 1 1 1 1 1 1 1
PSS 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1
PSUN 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
ROST 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
SCVL 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
SHOE 0 0 0 1 1 1 1 1 1 1 1 1 1 B
SJKI 0 0 0 0 1 1 1 1 1
SKFB 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
SMRT 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1
SSI 1 1 1 1 1 1
SYMS 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TJX 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TLB 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
TMAN 0 B 1 1 1 1 1 1
URBN 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1
URGI 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1
WALK 0 0 0 0 1 1 1 1 1 1 1 1 1 1
WTSL 1 1 1 1 1 1
ZUMZ 0 0 0 0 0 1 1 1 1
60

6.3 Financial Data
6.3.1 List of Financial Variables
The balance sheet, income statement and cash flow statement for each company were
obtained for every year historical values were available. All entries recorded from these
financial statements are listed in Table 6.4, though not all were used in analysis. Financial
data are reported in millions of dollars.
Table 6.4: Financial Items
Balance Sheet
- Cash
- Marketable Securities
- Accounts Receivables
- Inventories
- Other Current Assets
- Current Assets
- Long-Term Investments in
Securities
- Property, Plant and Equipment
- Accumulated Depreciation
- Amortizable Intangible Assets
- Goodwill and Non-
Amortizable Intangibles
- Other Non-Current Assets
- Total Assets
- Accounts Payable
- Notes Payable and Short-Term
Debt
- Current Maturities of Long-
Term Debt
- Other Current Liabilities
- Current Liabilities
- Long-Term Debt
- Deferred Taxes
- Other Non-Current Liabilities
- Total Liabilities
- Minority Interest in
Subsidiaries
- Preferred Stock
- Common Stock
- Retained Earnings
- Accumulated Other
Comprehensive Income
- Treasury Stock
- Shareholders Equity
- Total Liabilities and
Equities
Income Statement Cash Flow Statement
- Revenue
- Cost of Goods Sold
- Gross Profit
- Selling, General and
Administrative Expenses
- Other Operating Expenses
(Income)
- Operating Profit
- Interest Income
- Interest Expense
- Other Expenses (Income)
- Income Before Tax
- Income Tax Expense
- Income from Discontinued
Operations
- Extraordinary Gains (Losses)
- Changes in Accounting
Principles
- Net Income
- Cash Flow from
Operating Activities
- Cash Flow from
Investing Activities
- Cash Flow from
Financing Activities
- Change in Cash

6.3.2 Collection and Organization of Financial Variables
Because EDGAR does not provide its data in a manageable format, all financial data dating
back to 1988 (if available) were downloaded from the Capital IQ database accessible at the
Rotman School of Management. The Capital IQ database stores an assortment of data
10
for
each company in its own .xls (MS Excel) file. In this work, the most relevant information
was contained in the Balance Sheet, Income Statement and Cash Flow Statement
spreadsheets.

10
Capital IQ spreadsheets include: Key Stats, Income Statement, Balance Sheet, Cash Flow, Multiples, Historical
Capitalization, Capital Structure Summary, Capital Structure Details, Ratios, Supplemental, Industry Specific, Pension
OPEB, Segments

61

Companies within the same industry often label items in their financial statements
identically. However, in the instances where this was not the case, differences were resolved
in order for the comparison among companies to be legitimate. This required that financial
statements be converted into template statements (Figure 6.3). Templates were taken from
the Financial Statement Analysis Package (FSAP), Version 6.0, provided by Stickney et al.
(2007) (Figure 6.4 Figure 6.6). They are in .xls format and include the most important
items as well as unspecified fields (with the preceding label called Other) for the user to fit
his/her choice of data.
Creating template financial statements for all companies was time consuming yet
imperative. The conversion steps taken for one company typically involved:
1. creating a new company file where template statements were stored;
2. scouring through a companys original spreadsheets (in its Capital IQ file) and
extracting appropriate financial data to fill into the new template file; and,
3. validating the new template statements.
Macros coded in Visual Basic (VB) were written to automate the most tedious part of
the process (Step 2). After the templates were generated, they were manually inspected to
verify the numbers. For example, checkpoints were placed in the template statements to
ensure that the balance sheet, indeed, balanced and the computed net income was identical to
that stated in the income statement. If these checkpoints were violated, template data were
carefully reviewed and problems were manually fixed.

Figure 6.3: Creation of Template Financial Statements
AENA
ZUMZ
AEO
ANF
WTSL
WALK
AENA
ZUMZ
AEO
ANF
WTSL
WALK
Each companys financial
statements were retrieved
from Capital IQ, varying in
formats with different
labels for entries.
Each companys
financials were then
reorganized, conforming
to a template statement
format.
62

Figure 6.4: Example of Template Income Statement (in $M)

Figure 6.5: Example of Template Balance Sheet (in $M)
Year 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Revenues 587.6 832.1 1,093.5 1,371.9 1,382.9 1,435.4 1,889.6 2,322.0 2,794.4 3,055.4 2,988.9
Cost of Goods Sold -293.4 -398.4 -657.3 -824.5 -842.0 -885.9 -865.9 -1,244.2 -1,454.0 -1,632.3 -1,814.8
Gross Profit 294.2 433.7 436.2 547.4 541.0 549.5 1,023.7 1,077.7 1,340.4 1,423.1 1,174.1
Selling, General and Admin. Expense -198.5 -272.0 -266.5 -339.0 -328.2 -354.9 -591.9 -538.5 -665.6 -714.6 -734.0
Other Operating Expenses -8.6 -12.2 -23.2 -41.9 -53.4 -60.0 -68.2 -74.6 -81.9 -102.5 -126.3
Other Operating Income 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Non-Recurring Operating Gains 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Non-Recurring Operating Losses 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Operating Profit 87.1 149.5 146.6 166.5 159.4 134.7 363.6 464.6 592.9 606.0 313.8
Interest Income 2.4 0.0 0.0 0.0 0.0 2.0 0.0 0.0 0.0 39.3 18.9
Interest Expense 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Other Income or Gains 0.0 4.4 6.2 2.8 2.4 0.0 5.9 18.5 39.4 -1.3 0.0
Unusual Income or Gains 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Other Expenses or Losses 0.0 -4.6 0.0 0.0 -0.5 -1.4 -2.6 -6.1 -3.2 -7.7 -35.6
Unusual Expenses or Losses 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Income before Tax 89.5 149.4 152.8 169.2 161.3 135.3 366.8 477.0 629.1 636.4 297.0
Income Tax Expense -35.4 -58.7 -59.0 -63.8 -61.6 -52.2 -142.6 -183.3 -241.7 -236.4 -118.0
Minority Interest in Earnings 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Income from Discontinued Ops. 0.0 0.0 0.0 0.0 -11.5 -23.5 -10.9 0.4 0.0 0.0 0.0
Extraordinary Gains (Losses) 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Changes in Acctg. Principles 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
NET INCOME 54.1 90.7 93.8 105.5 88.1 59.6 213.3 294.2 387.4 400.0 179.1
Year 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Cash 71.9 76.6 133.4 180.4 194.5 137.1 219.4 130.5 59.7 116.1 473.3
Marketable Securities 13.4 91.9 27.9 45.1 47.0 200.7 370.2 621.0 754.1 503.9 10.5
Accounts Receivable 8.6 13.5 29.5 17.6 13.6 24.1 26.4 29.1 26.0 31.9 41.5
Inventories 49.7 60.4 84.1 91.1 124.7 120.6 170.6 210.7 263.6 286.5 294.9
Other Current Assets 11.0 20.2 43.8 42.5 46.8 48.2 78.8 85.4 85.6 82.5 105.1
CURRENT ASSETS 143.5 242.3 274.9 334.2 379.9 482.5 786.6 991.4 1,103.5 938.3 820.3
Long Term Investments in Securities 0.0 0.0 0.0 0.0 0.0 24.4 84.4 145.8 264.9 165.8 251.0
Property, Plant & Equipment - at cost 83.3 122.7 240.2 345.9 403.8 566.8 603.5 667.0 857.7 1,091.3 1,298.6
Accumulated Depreciation -29.9 -37.8 -56.9 -88.2 -136.3 -225.8 -263.6 -321.5 -376.0 -465.7 -558.4
Amortizable Intangible Assets 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Goodwill and Nonamort. Intangibles 0.0 0.0 26.6 24.0 23.6 10.1 10.0 10.0 10.0 11.5 10.7
Other Non-Current Assets 3.1 7.1 14.4 15.5 23.5 26.3 29.4 27.6 33.9 44.0 36.4
TOTAL ASSETS 196.9 327.3 484.9 615.9 671.0 858.0 1,220.8 1,492.6 1,860.0 1,741.2 1,822.2
Accounts Payable - Trade 18.6 30.7 42.0 39.1 50.6 71.3 108.9 139.2 171.2 157.9 152.1
Notes Payable and Short Term Debt 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 75.0
Current Maturities of Long Term Debt 30.8 39.1 52.4 61.4 45.7 59.9 81.1 100.6 115.9 111.7 94.1
Other Current Liabilities 10.4 18.7 54.7 50.7 45.3 77.7 92.6 111.7 177.6 106.6 80.6
CURRENT LIABILITIES 59.8 88.4 149.1 151.1 141.6 209.0 282.6 351.5 464.6 376.2 401.8
Long Term Debt 0.0 0.0 24.9 19.4 16.4 13.9 0.0 0.0 0.0 0.0 0.0
Deferred Taxes 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Other Non-Current Liabilities 0.0 1.7 1.3 1.4 5.9 72.2 82.8 98.6 97.6 151.0 152.9
TOTAL LIABILITIES 59.8 90.1 175.4 171.8 163.9 295.0 365.4 450.1 562.2 527.2 554.6
Minority Interest in Subsidiaries 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Preferred Stock 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Common Stock + Paid in Capital 65.0 89.7 119.4 152.0 155.6 157.5 269.8 372.2 455.9 495.9 516.1
Retained Earnings 89.9 180.5 274.3 379.8 468.5 522.3 726.8 978.9 1,302.3 1,601.8 1,694.2
Accum. Other Comprehensive Income -2.4 -5.7 -3.7 -4.8 -2.3 2.6 11.9 21.0 21.7 35.5 -14.4
Treasury Stock -1.3 0.0 -22.3 -24.9 -44.3 -45.0 -45.0 -216.5 -362.6 -792.7 -786.8
SHAREHOLDERS' EQUITY 151.2 264.5 367.7 502.1 577.5 637.4 963.5 1,155.6 1,417.3 1,340.5 1,409.0
TOTAL LIABILITIES AND EQUITIES 210.9 354.6 543.0 673.9 741.3 932.4 1,328.9 1,605.6 1,979.6 1,867.7 1,963.7
63

Figure 6.6: Example of Template Cash Flow Statement (in $M)
Since the data for each company were stored in a separate .xls file, retrieving a
common financial entry (e.g. cash) in a given year for all firms required traversing through
85 files. Because opening all files to extract specific data would be time consuming and
require substantial CPU power and memory, special VB macros were created with the ability
to:
1. retrieve data when the file was closed;
2. use query and store the data when the file was closed; and,
3. identify and manipulate the retrieved data into one open file.
Ultimately, these macros combined the balance sheet, income statement, and cash
flow statement of all 85 companies found across 85 files into one workable spreadsheet
(Figure 6.7). Each row represented a company and a year while each column of the
spreadsheet stored the values of a particular financial entry. There was a total of 994 DMUs
between 1988 and 2009; though only 701 DMUs were used between 1996 and 2009.
Year 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
NET INCOME 54.1 90.7 93.8 105.5 88.1 59.6 213.3 294.2 387.4 400.0 179.1
Add back Depreciation and Amortization Exps. 8.6 12.2 23.2 41.9 53.4 60.0 68.2 74.6 83.6 104.1 128.2
Other Addbacks to Net Income 2.8 5.8 7.0 3.1 -10.9 -9.3 59.9 45.1 48.6 42.1 55.2
Other Subtractions from Net Income 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
(Incr.) Decrease in Accts. Receivable - Trade -0.9 -4.9 -13.4 8.7 4.9 -9.3 3.9 10.5 7.4 -5.7 -10.1
(Incr.) Decrease in Inventories -13.4 -10.7 -5.6 -7.7 -34.5 14.5 -44.5 -39.1 -53.5 -19.1 -13.7
(Incr.) Decr. in Other Curr. Assets 1.7 4.3 5.4 4.5 15.2 2.7 7.4 10.1 11.6 -0.7 -11.4
Incr. (Decr.) in Acct. Payable - Trade -5.4 12.1 12.2 -2.1 12.8 13.4 23.2 29.4 32.3 -15.6 -3.1
Incr. (Decr.) in Other Current Liabilities 0.0 0.0 0.0 0.0 0.0 0.0 0.0 8.6 43.5 -31.4 -20.7
Incr. (Decr.) in Other Non-Current Liabs. 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Other Operating Cash Flows 12.3 23.4 28.1 21.1 -6.8 72.6 40.7 31.9 188.4 -9.5 -1.4
NET CF FROM OPERATIONS 59.8 132.9 150.6 174.9 122.1 204.0 372.0 465.2 749.3 464.3 302.2
Property, Plant, and Equipment Sold 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 12.3 0.0 0.0
Property, Plant, and Equipment Acquired -24.9 -45.6 -87.8 -119.3 -78.8 -77.5 -97.3 -81.5 -225.9 -250.4 -265.3
(Increase) Decrease in Marketable Securities -13.4 -85.4 66.5 -17.2 -30.3 -151.9 -178.4 -311.4 -437.4 354.2 344.9
Investments Sold 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Investments Acquired 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Other Investment Transactions 0.0 0.0 -88.1 2.0 -5.1 -2.9 5.4 -0.1 -0.1 -1.2 -1.2
NET CF FROM INVESTING ACTIVITIES -38.3 -130.9 -109.4 -134.5 -114.2 -232.3 -270.3 -393.1 -651.1 102.7 78.4
Increase in ST Borrowing 0.0 0.0 0.0 0.0 4.8 0.0 0.0 0.0 2.0 0.0 75.0
Decrease in ST Borrowing 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Increase in LT Borrowing 0.0 0.0 29.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
Decrease in LT Borrowing 0.0 0.0 -1.7 -5.7 -9.6 -5.4 -19.6 -0.7 -3.0 -1.9 -2.2
Issue of Capital Stock 2.0 2.7 10.2 15.8 1.8 1.1 57.5 48.2 28.4 13.2 3.8
Share repurchases - Treasury Stock 0.0 0.0 -22.3 -2.5 -19.5 -0.7 0.0 -171.5 -154.2 -450.6 -3.4
Dividend Payments 0.0 0.0 0.0 0.0 0.0 0.0 -8.8 -42.1 -61.5 -80.8 -82.4
Other Financing Transactions 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 19.5 6.2 0.7
NET CF FROM FINANCING ACTIVITIES 2.0 2.7 15.3 7.6 -22.4 -5.0 29.1 -166.1 -168.8 -514.0 -8.5
Effects of exchange rate changes on cash 0.0 0.0 0.4 -1.0 0.3 2.0 2.7 5.1 -0.2 3.4 -14.8
NET CHANGE IN CASH 23.6 4.6 56.9 47.0 -14.2 -31.3 133.5 -88.8 -70.8 56.3 357.3
64

Figure 6.7: Combining All Data into One Worksheet (in $M)

6.3.3 Distributions of Financial Variables
Figure 6.8 to Figure 6.31 present the distributions of selected variables with each count being
the average annual value for a given company (see List of Abbreviations for definitions of
variable acronyms). The mean, median and range are also provided. It is noted that a DMU
was predicted bankrupt up to 3 years prior to filing Chapter 11.

Figure 6.8: Histogram for Average Annual Cash
Mean = 4.7; Median = 0.5; Minimum = 0; Maximum = 98

Figure 6.9: Histogram for Average Annual AR

AENA
ZUMZ
AEO
ANF
WTSL
WALK
Balance Sheet Entries
Combine all data for
each company (stored
in own template file)
into onespreadsheet.
DMUs
Company Year Cash
Marketable
Securities
Accounts
Receivable Inventories
Other
Current
Assets
Current
Assets
AENA 2005 0.919977 0 0.150253 0.726366 0.198549 1.995145
AENA 2006 0.346759 0 1.943012 1.573487 0.041632 3.90489
AENA 2007 0.439238 0.027423 0.034754 1.753363 0.208955 2.463733
AENA 2008 0.695066 0 0.248694 2.088256 0.145459 3.177475
AEO 1993 0 0 0 0 0 0
AEO 1994 0.231 0 5.365 45.164 9.151 59.911
AEO 1995 0.542 0 7.255 79.019 9 95.816
AEO 1996 19.986 0 5.642 23.394 7.32 56.342
AEO 1997 34.326 0 3.556 27.117 8.761 73.76
AEO 1998 48.359 0 7.647 36.278 10.189 102.473
AEO 1999 71.94 13.36 8.56 49.688 10.956 154.504
AEO 2000 76.581 91.911 13.471 60.375 20.224 262.562

ZUMZ 2004 0.578 0 1.039 20.802 1.063 23.482
ZUMZ 2005 1.026 0 1.911 23.23 2.025 28.192
ZUMZ 2006 4.737 38.264 3.746 30.559 1.649 78.955
ZUMZ 2007 8.161 43.816 5.223 42.157 5.144 104.501
ZUMZ 2008 11.945 64.587 4.775 48.721 5.529 135.557
ZUMZ 2009 33.057 45.525 4.555 51.974 8.202 143.313
0
5
10
15
20
25
30
35
0
0
.
2
5
0
.
5
0
.
7
51
1
.
2
5
1
.
5
1
.
7
52
2
.
2
5
2
.
5
2
.
7
53
3
.
2
5
3
.
5
3
.
7
54
4
.
2
5
4
.
5
4
.
7
55
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
M
o
r
e
Bin in $M
Active Bankrupt
65

Figure 6.10: Histogram for Average Annual Inv

Figure 6.11: Histogram for Average Annual CA
Mean = 22; Median = 2.9; Minimum = 0; Maximum = 510

Figure 6.12: Histogram for Average Annual PPE

Figure 6.13: Histogram for Average Annual AD

Figure 6.14: Histogram for Average Annual TA

Figure 6.15: Histogram for Average Annual AP

Figure 6.16: Histogram for Average Annual CM

Figure 6.17: Histogram for Average Annual CL
0
5
10
15
20
0
0
.
51
1
.
52
2
.
53
3
.
54
4
.
55
5
.
56
6
.
57
7
.
58
8
.
59
9
.
5
1
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
2
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
2
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0
0
.
51
1
.
52
2
.
53
3
.
54
4
.
55
5
.
56
6
.
57
7
.
58
8
.
59
9
.
5
1
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
02468
1
0
1
2
1
4
1
6
1
8
2
0
2
2
2
4
2
6
2
8
3
0
3
2
3
4
3
6
3
8
4
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0
0
.
2
0
.
4
0
.
6
0
.
81
1
.
2
1
.
4
1
.
6
1
.
82
2
.
2
2
.
4
2
.
6
2
.
83
3
.
2
3
.
4
3
.
6
3
.
84
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0
0
.
2
0
.
4
0
.
6
0
.
81
1
.
2
1
.
4
1
.
6
1
.
82
2
.
2
2
.
4
2
.
6
2
.
83
3
.
2
3
.
4
3
.
6
3
.
84
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0
0
.
51
1
.
52
2
.
53
3
.
54
4
.
55
5
.
56
6
.
57
7
.
58
8
.
59
9
.
5
1
0
M
o
r
e
Bin in $M
Active Bankrupt
66

Figure 6.18: Histogram for Average Annual LTD

Figure 6.19: Histogram for Average Annual TL

Figure 6.20: Histogram for Average Annual RE
Mean = 19; Median = 1.0; Minimum = -7.7; Maximum = 390

Figure 6.21: Histogram for Average Annual SE
Mean = 25; Median = 3.3; Minimum = -0.4; Maximum = 560

Figure 6.22: Histogram for Average Annual Rev
Mean = 75; Median = 11; Minimum = 0; Maximum = 1800

Figure 6.23: Histogram for Average Annual
COGS

Figure 6.24: Histogram for Average Annual SGA

Figure 6.25: Histogram for Average Annual II
Mean = 0.2; Median = 0; Minimum = 0; Maximum = 5.6
0
5
10
15
20
25
30
0
0
.
2
0
.
4
0
.
6
0
.
81
1
.
2
1
.
4
1
.
6
1
.
82
2
.
2
2
.
4
2
.
6
2
.
83
3
.
2
3
.
4
3
.
6
3
.
84
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
2
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
-
10123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
2
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
0123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
1
8
1
9
2
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
05
1
0
1
5
2
0
2
5
3
0
3
5
4
0
4
5
5
0
5
5
6
0
6
5
7
0
7
5
8
0
8
5
9
0
9
5
1
0
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
40
05
1
0
1
5
2
0
2
5
3
0
3
5
4
0
4
5
5
0
5
5
6
0
6
5
7
0
7
5
8
0
8
5
9
0
9
5
1
0
0
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
02468
1
0
1
2
1
4
1
6
1
8
2
0
2
2
2
4
2
6
2
8
3
0
3
2
3
4
3
6
3
8
4
0
M
o
r
e
Bin in $M
Active Bankrupt
0
10
20
30
40
50
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
M
o
r
e
Bin in $M
Active Bankrupt
67

Figure 6.26: Histogram for Average Annual IE
Mean = 0.2; Median = 0; Minimum = 0; Maximum = 5.6

Figure 6.27: Histogram for Average Annual NI
Mean 4.2; Median = 0.2; Minimum = -4.7; Maximum = 98

Figure 6.28: Histogram for Average Annual CFO
Mean = 7.3; Median = 0.6; Minimum = -3.2; Maximum = 150

Figure 6.29: Histogram for Average Annual CFF
Mean = -6.1; Median = -0.5; Minimum = -200; Maximum = 0.2

Figure 6.30: Histogram for Average CFF
Mean = -1.1; Median = 0; Minimum = -38; Maximum = 28

Figure 6.31: Histogram for Average Annual CC
Mean = 0.1; Median = 0; Minimum = -50; Maximum = 24

6.3.4 Correlations of Financial Variables
Correlations between the values of all financial accounts in the balance sheet, income
statement and cash flow statements were found. Table 6.5 to Table 6.7 present those
correlations greater than 0.85 and significant at the 0.01 level.

0
10
20
30
40
50
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
0
0
.
1
0
.
2
0
.
3
0
.
4
0
.
5
0
.
6
0
.
7
0
.
8
0
.
91
1
.
1
1
.
2
1
.
3
1
.
4
1
.
5
1
.
6
1
.
7
1
.
8
1
.
92
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
40
-
3
-
2
-
10123456789
1
0
1
1
1
2
1
3
1
4
1
5
1
6
1
7
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
-
4
-
3
.
7
5
-
3
.
5
-
3
.
2
5
-
3
-
2
.
7
5
-
2
.
5
-
2
.
2
5
-
2
-
1
.
7
5
-
1
.
5
-
1
.
2
5
-
1
-
0
.
7
5
-
0
.
5
-
0
.
2
50
0
.
2
5
0
.
5
0
.
7
51
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
40
45
-
0
.
1
-
0
.
0
50
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
M
o
r
e
Bin in $M
Active Bankrupt
0
5
10
15
20
25
30
35
-
1
-
0
.
9
-
0
.
8
-
0
.
7
-
0
.
6
-
0
.
5
-
0
.
4
-
0
.
3
-
0
.
2
-
0
.
10
0
.
1
0
.
2
0
.
3
0
.
4
0
.
5
0
.
6
0
.
7
0
.
8
0
.
91
M
o
r
e
Bin in $M
Active Bankrupt
68

Table 6.5: Highly Correlated Financial Accounts on the Balance Sheet
Refer to List of Abbreviations for definitions of variable acronyms.
Accounts Correlation Accounts Correlation
Inv, CA 0.95 PPE, CM 0.87
Inv, PPE 0.87 PPE, CL 0.89
Inv, Other NonCA 0.91 PPE, TL 0.88
Inv, TA* 0.94 PPE, SE 0.88
Inv, AP 0.94 AD, Other NonCA 0.93
Inv, CM 0.87 AD, TA* 0.90
Inv, CL 0.94 AD, CM 0.85
Inv, TL 0.92 AD, CL 0.86
Inv, SE 0.86 AD, TL 0.85
OtherCA, CA 0.88 Other NonCA, TA* 0.96
OtherCA, PPE 0.85 Other NonCA, AP 0.89
OtherCA, Other NonCA 0.89 Other NonCA, CM 0.88
OtherCA, TA* 0.90 Other NonCA, CL 0.92
OtherCA, CL 0.86 Other NonCA, TL 0.91
OtherCA, TL 0.87 Other NonCA, SE 0.91
CA, PPE 0.91 TA*, AP 0.92
CA, AD 0.89 TA*, CM 0.91
CA, Other NonCA 0.93 TA*, CL 0.95
CA, TA* 0.98 TA*, TL 0.95
CA, AP 0.93 TA*, SE 0.94
CA, CM 0.90 AP, CM 0.87
CA, CL 0.95 AP, CL 0.96
CA, TL 0.93 AP, TL 0.91
CA, SE 0.93 CM, CL 0.94
PPE, AD 0.98 CM, TL 0.93
PPE, Other NonCA 0.97 CL, TL 0.97
PPE, TA* 0.93 CL, SE 0.86
PPE, AP 0.87 RE, SE 0.86

*Total Assets = Total Liabilities and Shareholders Equity
Table 6.6: Highly Correlated Financial Accounts on the Income Statement
Accounts Correlation
Revenue, CoGS 0.99
Revenue, Gross Profit 0.98
Revenue, SG&A 0.98
CoGS, Gross Profit 0.94
CoGS, SG&A 0.95
Gross Profit, SG&A 0.99
Operating Profit, Income before Tax 0.94
Operating Profit, Income Tax Expense 0.91
Operating Profit, Net Income 0.89
Income before Tax, Income Tax Expense 0.94
Income before Tax, Net Income 0.95
Income Tax Expense, Net Income 0.87

On the Balance Sheet (Table 6.5), the strongest correlations were between larger
accounts such as total current assets, total assets, total current liabilities and total liabilities,
etc., which are sums of other entries. Smaller accounts such as cash, marketable securities,
69

accounts receivable, long-term investments in securities, amortizable intangible assets,
goodwill assets that cannot be amortized, notes payable and short-term debt, deferred taxes,
minority interest in subsidiaries, preferred stock, common stock, treasury stock and retained
earnings, were weakly correlated with other financial accounts. The strongest correlations in
the Income Statement (Table 6.6) were also between its larger accounts. And in the Cash
Flow Statement, only weak correlations existed between the cash flows (operating, investing
and financing) and the overall change in cash.
As for correlations across the financial statements (Table 6.7), there were many
highly correlated accounts between the Balance Sheet and Income Statement. This is
attributed to the redundancy of revenue, CoGS, Gross Profit and SG&A (as they were highly
correlated among themselves), and their inherent connection to assets and liabilities. When
correlating the Balance Sheet and Income Statement with the Cash Flow Statement, only two
strong relationships emerged: Gross Profit and Operating Cash Flow, and Operating Profit
and Operating Cash Flow.
Generally, a weak correlation implies that variables are independent while a strong
correlation suggests that trimming the dataset is desirable.
Table 6.7: Highly Correlated Financial Accounts across Statements
Accounts Correlation Accounts Correlation
Inventories, Revenue 0.86 Total Assets*, Gross Profit 0.89
Inventories, CoGS 0.86 Total Assets*, SG&A 0.89
Inventories, Gross Profit 0.84 Accounts Payable, Revenue 0.86
Inventories, SG&A 0.85 Accounts Payable, CoGS 0.86
Current Assets, Revenue 0.89 Accounts Payable, Gross Profit 0.84
Current Assets, CoGS 0.87 Current Liabilities, Revenue 0.89
Current Assets, Gross Profit 0.89 Current Liabilities, CoGS 0.88
Current Assets, SG&A 0.88 Current Liabilities, Gross Profit 0.87
PPE, Gross Profit 0.85 Current Liabilities, SG&A 0.88
PPE, SG&A 0.85 Total Liabilities, Revenue 0.87
Other Non-Current Assets, Revenue 0.88 Total Liabilities, CoGS 0.85
Other Non-Current Assets, CoGS 0.85 Total Liabilities, Gross Profit 0.86
Other Non-Current Assets, Gross Profit 0.88 Total Liabilities, SG&A 0.87
Other Non-Current Assets, SG&A 0.87 Gross Profit, Operating Cash Flow 0.85
Total Assets*, Revenue 0.89 Operating Profit, Operating Cash Flow 0.85
Total Assets*, CoGS 0.87

*Total Assets = Total Liabilities and Shareholders Equity

6.3.5 Profitability Analysis
Profitability ratios return on assets (ROA), profit margin (PM) and total assets turnover
(TAT) were calculated for each company for each year with available data. Median ROA,
70

PM and TAT for each company (across all years) were then computed and correlated with its
state. The median was taken to eliminate outliers and because the distributions of these ratios
were not normally distributed. As expected, ROA of active companies was higher than
bankrupt firms (Table 6.8). Similar trends can be observed for PM and TAT. With these
data, industry analysis was conducted by plotting median TAT and median PM for all
companies (Figure 6.32).
Table 6.8: Average Median ROAs, PMs, TATs by State
A company was classified as active if it never filed Chapter 11 and as bankrupt if it filed at least once.
Ratio Industry Average Active Bankrupt All
TAT 2.1 2.4 0.9 2.7 0.9 2.4 0.9
PM 3.0% 4.2% 3.0% -3.1% 6.0% 2.1% 5.5%
ROA 12% 10.6% 6.0% -3.3% 14.7% 6.9% 10.9%

Figure 6.32: Median Efficiency and Profitability Chart for All Companies
Recall that in Table 2.4, companies in the retail apparel industry should be classified
as Type C because they operate in a highly competitive environment. The strategic focus
for ROA for Type C companies should thus be on increasing assets turnover. However, in
this sample (Figure 6.32), firms were more clearly distinguishable by profit margin: those
active had higher profit margins compared to those that filed for bankruptcy. Hence, while
these ratios provided hints of a companys health, they alone were unreliable for use as
benchmarks or measures of financial health as seen by the scatter (variance) in the data,
which can be attributed to taking the median between all years (and thereby ignoring possible
-15%
-10%
-5%
0%
5%
10%
15%
0.0 1.0 2.0 3.0 4.0 5.0
Profit
Margin
Assets Turnover
Active Bankrupt Industry Average
71

market and economic influences that produce outliers). It is also noted that in Table 6.8,
the product of PM and TAT do not necessarily match up with ROA because in some
instances, the denominator of a ratio was zero.
To further decipher profit margin and total asset turnover, several ratios were
calculated for each company for each year: Cost of Goods Sold (COGS) %, Selling, General
and Administrative Expenses (SGA) % and Income Tax Expense % as well as average
accounts receivable turnover, average inventory turnover and average fixed assets turnover.
The averages of these ratios were then correlated to company state (Table 6.9 and Table
6.10). As expected, CoGS% and SG&A% were lower for active firms because they likely
learned to be more operationally efficient. With respect to turnover (TO), the accounts
receivable and inventory TOs averages were higher for active companies while fixed assets
TO was the same across both states.
Table 6.9: Averages of Ratios in Profit Margin Breakdown
A DMU was classified bankrupt up to 3 years prior to filing Chapter 11.
State COGS % SGA % Income Tax Expense %
Active 62% 10% 31% 18% 3% 2%
Bankrupt 66% 9% 32% 11% 2% 4%
All 63% 10% 31% 17% 3% 3%

Table 6.10: Averages of Ratios in Assets Turnover (TO) Breakdown
State Accounts Receivable TO Inventory TO Fixed Assets TO
Active 150 220 100 290 4 3
Bankrupt 96 95 15 20 4 2
All 147 210 94 280 4 3

Return on Common Equity (ROCE) was also investigated (Table 6.11). As expected,
the ROCE of active companies was greater than that of bankrupt firms.
Table 6.11: Average of Median ROCE
A company was classified as active if it never filed Chapter 11 and as bankrupt if it has filed at least once.
State Average Median ROCE
Active 15.9 % 13.7%
Bankrupt -7.3% 20.9%
All 9.4% 19.1%
Moreover, the ROCE of each company was compared to its ROA to determine leverage
(Table 6.12) with those missing ROA or ROCE values removed from the dataset. Active
firms were more likely positively leveraged, while approximately half the bankrupt firms
were positively leveraged, making this an unreliable index for determining bankruptcy.

72

Table 6.12: Company Leverage Position
State Total Negatively Leveraged Positively Leveraged
Active 55 4 (7%) 51 (93%)
Bankrupt 19 10 (53%) 9 (47%)

6.3.6 Short-Term Liquidity and Long-Term Solvency
Ratios for liquidity and solvency (defined in Table 2.5) were computed for each company for
every year and then correlated to its state. As expected, current ratio, quick ratio and
operating cash flow to current liabilities ratio were higher for active firms than those
bankrupt (Table 6.13). The days outstanding for receivables and inventory were lower for
active companies; though, for days revenue in cash, the ratio was higher (perhaps because
bankrupt firms need cash immediately to pay bills) (Table 6.14). As for solvency, the
interest coverage ratio was highest for active companies (i.e. they have sufficient income to
pay interest) who also had lower liabilities/long-term debt ratios (Table 6.15).
Table 6.13: Average Liquidity Ratios
State Current Ratio Quick Ratio CFO to CL
Active 2.5 1.1 1.1 1.0 0.6 0.6
Bankrupt 1.8 1.0 0.6 0.6 0.2 0.5
All 2.4 1.1 1.1 0.9 0.6 0.6
Table 6.14: Average Liquidity Ratios
State Average Collection Period Average Inventory Period Days Revenue in Cash
Active 9 16 58 56 35 66
Bankrupt 10 10 80 71 12 15
All 9 15 60 58 33 64
Table 6.15: Average Solvency Ratios
State
Interest
Coverage
Liabilities
to Assets
Liabilities to
Shareholders Equity
Long-Term Debt
to Capital
Long-Term Debt to
Shareholders Equity
Active 41.7 79.1 0.4 0.2 0.9 1.2 0.2 0.3 0.2 0.5
Bankrupt 9.1 45.4 0.7 0.3 1.3 3.0 0.2 0.4 0.3 1.3
All 38.6 77.1 0.5 0.2 0.9 1.4 0.2 0.3 0.2 0.6

While there were clear trends between liquidity and solvency ratios, and state, no
information on industry averages was readily available for these ratios; thus, the relevance of
these ratios (determined by percentage of error) as proxies for company health was qualified
by creating a set of bankruptcy thresholds and/or boundaries based on the dataset. In other
words, to determine how well a given ratio R predicted bankruptcy, a value x was found such
that when x > R, a company was classified as active and when x < R a firm was classified as
73

bankrupt (Figure 6.33). Another objective of x was to classify as many companies correctly.
With this x, counts for true positive, true negative, false positive and false negative were
computed to determine if the given ratio was a good indicator of health.
Interestingly, with this dataset, the optimal x (i.e. with minimum Type I and II Errors)
for all ratios was always equal to the minimum value of R; there was no such x better than
always predicting active. This is likely due to the large percentage of active companies in
the dataset. Thus, with such a primitive predictive approach, liquidity and solvency ratios
and company were unreliable measures of company state. Perhaps instead at the intersection
of error, there may be an x such that when x < R, a company could be classified as bankrupt;
and a y such that when y > R, a company could be classified as active; while the range x y
could be indeterminate like Altmans Z-Score (Section 2.3.1).

Figure 6.33: Strategy for Deriving a Threshold for Evaluating Ratios

6.3.7 Earnings Manipulation Risk
The Beneish Index (Section 2.4.1) was calculated for all companies with a particular focus on
bankrupt firms. Table 6.16 presents average scores by company state. Surprisingly,
bankrupt companies were no more likely to engage in manipulation despite what ones
intuition would suspect (
Table 6.17). This is likely because Beneish based his model on manufacturing firms.

Value of Ratio R
x
Subset of
Active
Companies
Subset of
Bankrupt
Companies
Type I
Error
Type II
Error
74

Table 6.16: Average Beneish Index
State Beneish
Active 0.05 0.17
Bankrupt 0.02 0.30
All 0.05 0.19

Table 6.17: Bankrupt Companies and Earnings Manipulation Risk
Year 0 Year 1 Year 2 Year 3
Classified Suspicious 2 (17%) 2 (17%) 2 (17%) 3 (25%)
Classified Non-Suspicious 10 (83%) 10 (83%) 10 (83%) 9 (75%)

6.3.8 Quality of Income
High-quality earnings have a high correlation with cash flows, which are more difficult to
manipulate than reported income. Table 6.18 shows the average values of Quality of Income
by state. As expected, the average of active companies was higher than 1 (implying high
earnings quality) whereas for bankrupt companies, it was below 1 (implying a red flag). The
quality of income measure was also evaluated as a determinant of bankrupt. While it yielded
a true positive rate of 96%, it produced a false positive rate of 83%, making it an inadequate
proxy.
Table 6.18: Quality of Income
State Quality of Income
Active 1.3 3.6
Bankrupt 0.4 2.7
All 1.2 3.5

6.4 Managerial Decision-Making Data
Of the 85 firms studied, annual reports for all but two companies (BSST and CSS) were
available from EDGAR, SEC and/or company websites. Because of the abundance of
information in the Notes, MD&A, and Auditors Report, a comprehensive review of fraud
literature and the retail-apparel industry was conducted to determine what was worth
collecting. A particular variable was chosen only if: 1. it was relevant to the retail-apparel
industry; 2. it was a commonly reported item across all companies; 3. it was easily extracted
from the annual report; and 4. it could be translated into a useful scale (i.e. binary). Eight
types of information were extracted from each annual report for each year for each company:
75

1. Significant Related-Party Transactions, in the form of leases or loans to the
executive, etc.;
2. Auditors Opinion, either unqualified or qualified;
3. Independent Auditing Company, typically either Arthur Andersen LLP, Deloitte &
Touche LLP, Ernst & Young LLP, KPMG Peat Marwick LLP or
PricewaterhouseCoopers LLP;
4. Legal proceedings, such as those arising in the normal course of business that do not
have a material adverse effect on the company, litigations that lead to significant
payouts and filing of Chapter 11 bankruptcy;
5. Name of Chairman;
6. Name of Chief Executive Officer;
7. Name of Chief Financial Officer; and,
8. Retirement plans, where if applicable, employees are eligible to participate in the
companys 401(k) plan or there is a specific company-sponsored pension program.
Because each annual report is presented as a .pdf and/or .doc file, data collection required
manual key word searches (e.g. related party transactions). Relevant information was
recorded as categorical variables in an .xls file in the same format as financial data (Figure
6.34). This was an extremely tedious process as approximately 1000 annual reports were
read. It is also noted that annual reports prior to 1994 (before EDGARs mandatory filing)
could not be found and not all were available post-1996. The categorical variables were then
translated to have numerical values associated with their possible outcomes (Table 6.19). All
variables except for Legal Proceedings were binary and assigned a value of 0 or 1. For
Legal Proceedings, the outcomes None, Insignificant, Significant, Going Concern
and Bankruptcy translated to 0, 2, 10, 20 and 25, respectively, based on the frequency of
their occurrence among all DMUs.

76

Figure 6.34: Example of Organization of Qualitative Information Extracted from Annual Reports
Table 6.19: Managerial Decision-Making Variables
Variable Outcome Value Assigned
Significant Related Party Transactions
None 0
Yes 1
Auditors Opinion
Unqualified 0
Qualified 1
Legal Proceedings
None 0
Insignificant 2
Significant 10
Going Concern 20
Bankruptcy Filing 25
Retirement Plan
None 0
Yes 1
Auditor Change
(Change in auditor company)
None 0
Yes 1
Turnover of Management
(Change in either Chairman, Chief Executive Officer
or Chief Financial Officer)
None 0
Yes
1, 2 or 3 (depending on how many
changed in that year)

As shown in Table 6.20, legal proceedings, auditor change and management turnover
had a statistically significant relationship to a companys state. Specifically, an unhealthy
firm was more likely to make changes to their management (Table 6.21) and auditors (Table
6.22), and be involved in more serious legal proceedings (Table 6.23).
Company Year
Related
Parties
Legal
Proceedings Auditor Opinion
Chairman of the
Board
Chief Executive
Officer
Chief Financal
Officer
AENA 2005
AENA 2006
AENA 2007 Yes Insignificant Goldstein Schechter Koch Unqualified Ilia Lekach Ilia Lekach Jeffrey Geller
AENA 2008
AEO 1993
AEO 1994
AEO 1995
AEO 1996
AEO 1997 Yes None Ernst & Young LLP Unqualified Jay Schottenstein Jay Schottenstein Laura Weil
AEO 2003 Yes Insignificant Ernst & Young LLP Unqualified Jay Schottenstein James O'Donell Laura Weil
AEO 2006 Yes Insignificant Ernst & Young LLP Unqualified Jay Schottenstein James O'Donell Joan Holstein Hilson

ZUMZ 2006 Yes None PricewaterhouseCoopers LLP Unqualified Tom Campion Rick Brooks Brenda Morris
ZUMZ 2007 Yes None Pricewaterhouse Coopers LLP Unqualified Tom Campion Rick Brooks Brenda Morris
ZUMZ 2008 Yes None Moss Adams LLP Unqualified Tom Campion Rick Brooks Trevor Lang
ZUMZ 2009 Yes None Moss Adams LLP Unqualified Tom Campion Rick Brooks Trevor Lang
77

Table 6.20: Chi-Square Test Results of Managerial Decision-Making Data and State
Variable State
2
df p
Related Party Transactions 8.8 4 0.067
Auditors Opinion 0.5 4 0.969
Legal Proceedings 226.1 16 0.000
Retirement Plan 3.7 8 0.884
Auditor Change 20.3 4 0.000
Management Turnover 14.7 4 0.005

Table 6.21: Likelihood of Management Turnover
State Average Number of Management Changes Times Over Average
Active 0.4 0.9
Bankrupt 0.8 1.9
All 0.4 -

Table 6.22: Probability of Auditor Change by State
State Probability of Auditor Change
Active 0.05
Bankrupt 0.10
All 0.05

Table 6.23: Percentage of DMUs with Non-Serious and Serious Legal Proceedings by State
State Legal Proceedings* Count Percentage
Active
Non-serious 606 93.1%
Serious 45 6.9%
Bankrupt
Non-serious 33 66.0%
Serious 17 34.0%
*Non-serious legal proceedings include the categories: None and Insignificant;
Serious legal proceedings include the categories: Significant, Going Concern and Bankruptcy

6.5 Outliers and Missing Data
Because numerical data were collected directly from financial statements prepared by
companies and approved by independent auditors, measurement error was assumed to be
negligible as one cannot be concerned about observations that are not in ones control.
However, this did not rule out the outliers which appear with the conversion of values from
absolute to relative terms. In some instances, there were some extreme percentage changes
from year-to-year. These occurred when there was a dramatic increase or decrease of a
variable from the previous year (since % change =
Value at Year
t
- Value at Year
t-1
Value at Year
t-1
100%) and were
addressed by rounding down or up to a reasonable level (like one magnitude above or below
the average). For example, if a company with minimal to no long-term debt (i.e. $100,000)
suddenly took a large loan (i.e. $10,000,000), the change in total liabilities from the previous
78

year would be 9900%. In this case, the change was scaled to 300% (20 times over the
average). Dealing with data in this manner recognized the value of and reflected upon the
extreme change relative to other DMUs (without removing it completely) and did not skew
the rest of the data.
Because annual reports are professionally audited, there were no missing financial
data entries with any of them. However, in cases where an annual report could not be found
for a given company for a given year (and thus, no data on managerial decision-making data
could be collected), the corresponding DMU was removed entirely.

6.6 Economic Data
The historical values for over 100 economic factors were collected from federal agencies
with statistical programs such as the Bureau of Economic Analysis (Department of
Commerce), the Bureau of Labour Statistics (Department of Labour), the Census Bureau
(Department of Commerce), National Agricultural Statistics Service (Department of
Agriculture) and the National Retail Federation. These factors fall into seven broad
categories (Table D.1) each having one of two relationships to the economy either:
Procyclic, if it moves in the same direction as the economy; or,
Countercyclic, if it moves in the opposite direction as the economy.
Each factor can also be classified by time, either:
Leading, if it changes before economy changes;
Lagging, if it does not change direction until a few quarters after the economy
does; or,
Coincident, if it moves with the economy.
As the majority of the factors were strongly correlated with one another (i.e. greater
than 0.9), this prompted a reduction to a manageable set. Ultimately, 17 economic factors
having a direct effect on the apparel industry were considered (Table 6.24) with information
stored in .xls format (Figure 6.35). For the most part, these factors were weakly correlated (<
|0.4|) with one another (Table 6.25).

79

Table 6.24: Economic Factors in this Thesis
Factor
Measured
in
Analysis
in
Timing and Relation to
Business Cycle
General Economic Factors
E1 GDP Rate $B % Coincident, Procyclical
E2 Debt as % of GDP % Same Coincident, Countercyclical
E3 Inflation % Same Coincident, Procyclical
E4 Interest rate % Same Coincident, Procyclical
E5 Unemployment Rate % Same Lagged, Countercyclical
Apparel Factors
E6
Personal Consumption Expenditures: Clothing
& Footwear
$B % Coincident, Procyclical
E7 GDP: Clothing & Footwear $B % Coincident, Procyclical
E8 CPI: Apparel
Index
(1982)
% Coincident, Procyclical
E9 Industrial Production: Clothing
Index
(2007)
% Coincident, Procyclical
E10
Apparel Unit Labor Cost = (total labour
compensation / hours) / productivity
% Same Coincident, Procyclical
E11 Apparel Labor Productivity = output/hours % Same Coincident, Procyclical
E12 Apparel Imports $ % Coincident, Procyclical
E13 Apparel Exports $ % Coincident, Countercyclical
Other Factors
E14 New Privately Owned Housing Units Started 1000s Same Leading, Procyclical
E15 Median Number of Months for a Sale Months Same Leading, Countercyclical
E16 Oil price $/bbl Same Coincident, Countercyclical
E17 Cotton price cent/lb Same Coincident, Countercyclical

Figure 6.35: Example of Organization of Economic Data
Year GDP Growth Rate (%) Debt as % of GDP Inflation (%) Interest Rate (%) Unemployment Rate (%)
1988 7.69 51.02 4.08 7.57 5.5
1989 7.48 52.12 4.84 9.21 5.3
1990 5.81 55.74 5.39 8.10 5.6

2007 5.07 63.99 2.85 5.02 4.6
2008 2.58 69.15 3.85 1.92 5.8
2009 -1.28 83.29 -0.34 0.16 9.3
80

Table 6.25: Spearman Correlation Coefficients of Economic Factors

*Codes correspond to Table 6.24; significant at the 0.01 level (2-tailed)

6.7 Market and Stock Performance
The 85 public firms in this thesis were either traded on the NASDAQ, NYSE or OTC (Figure
6.36). Therefore, the NASDAQ and NYSE Composite Indices were analyzed as measures
that reflect overall stock performance. The NASDAQ Composite is a stock market index of
common stocks listed on the NASDAQ stock market with over 3000 components. Similarly,
the NYSE Composite is a stock market index of common stocks listed on the New York
Stock Exchange with over 2000 components. Moreover, the S&P Composite was considered
as it combines three indices (the S&P 500, the S&P MidCap 400 and the S&P SmallCap 600)
that cover approximately 90% of the U.S. Market Capitalization. It is designed for investors
seeking to replicate the performance of the U.S. equity market or benchmark against a
representative universe of tradable stocks. Figure 6.37 plots the average yearly returns of
these composites, which are leading, procyclical and highly correlated with one another
(Table 6.26).
E1 E2 E3 E4 E5 E6 E7 E8 E9 E10 E11 E12 E13 E14 E15 E16 E17
E1 1.00 -0.17 0.07 0.16 -0.45 0.88 0.86 0.24 0.28 -0.30 0.17 0.54 0.16 0.50 -0.26 0.03 -0.03
E2 1.00 0.30 0.26 0.28 -0.22 -0.09 0.69 0.30 0.06 -0.05 0.01 0.60 -0.57 0.71 0.20 0.61
E3 1.00 0.09 -0.03 0.03 0.02 0.16 -0.06 -0.06 -0.06 -0.26 0.23 -0.22 0.19 0.78 0.00
E4 1.00 -0.60 0.40 0.09 0.34 0.65 -0.26 0.31 0.53 0.65 -0.57 0.52 -0.36 0.56
E5 1.00 -0.70 -0.52 -0.21 -0.20 0.06 -0.10 -0.25 -0.01 0.04 -0.23 0.17 0.11
E6 1.00 0.86 0.19 0.42 -0.46 0.34 0.57 0.24 0.35 -0.06 -0.17 0.06
E7 1.00 0.28 0.21 -0.27 0.13 0.33 0.05 0.58 -0.13 0.00 -0.05
E8 1.00 0.28 0.17 -0.17 0.20 0.47 -0.26 0.55 0.10 0.31
E9 1.00 -0.74 0.80 0.69 0.85 -0.25 0.19 -0.55 0.66
E10 1.00 -0.88 -0.45 -0.51 -0.07 0.17 0.37 -0.42
E11 1.00 0.44 0.49 -0.07 -0.06 -0.47 0.44
E12 1.00 0.58 0.09 -0.13 -0.45 0.58
E13 1.00 -0.45 0.40 -0.22 0.72
E14 1.00 -0.80 0.03 -0.45
E15 1.00 0.03 0.36
E16 1.00 -0.22
E17 1.00
81

Figure 6.36: Count and Percentage of Companies Traded on Specific Exchanges

Figure 6.37: Average Yearly Returns of Composite Indices
Table 6.26: Correlation between Composite Indices
Significant at the 0.01 level (2-tailed)

NASDAQ NYSE S&P
NASDAQ 1.00 0.82 0.88
NYSE

1.00 0.95
S&P

1.00

In addition to indices reflecting the general market performance, end-of-month stock prices
were collected for each company (except CSS, DUSS, EXLA since information was not
available) between 1988 and 2009 from the CRSP database as well as Google and Yahoo!
Finance websites. The monthly prices were used to calculate monthly returns, from which
annual return, annual volatility and annual beta
11
were computed. Table 6.27 presents the
average annual return, volatility and beta by state while Figure 6.38, Figure 6.39 and Figure

11
Volatility is equivalent to standard deviation while beta is covariance of company return and market return, divided by the
variance of the market (Section 2.5). It is standard practice to use the S&P Composite to represent market performance.
NASDAQ,
38, 45%
NYSE, 30,
35%
OTC, 17,
20%
-5.0
-4.0
-3.0
-2.0
-1.0
0.0
1.0
2.0
3.0
-60
-40
-20
0
20
40
60
80
100
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
Annual Return
for S&P
(%)
Annual Return
for NASDAQ
and NYSE
(%)
Year
NASDAQ Composite Index NYSE Composite Index S&P Composite Index
82

6.40 plot these measures by year (from 1994 to 2009). As expected, healthier firms had
higher returns, and are less volatile with lower beta.
Table 6.27: Average Annual Return, Volatility and Beta
State Return Volatility Beta
Active 0.013 0.055 0.16 0.08 1.16 1.45
Bankrupt -0.019 0.099 0.26 0.15 1.63 3.11
All 0.011 0.059 0.17 0.09 1.19 1.61

Figure 6.38: Average Return

Figure 6.39: Average Volatility

-0.4
-0.3
-0.2
-0.1
0.0
0.1
1994 1996 1998 2000 2002 2004 2006 2008
Average
Return
(%)
Year
Active Bankrupt
0.0
0.1
0.2
0.3
0.4
0.5
1994 1996 1998 2000 2002 2004 2006 2008
Average
Volatility
(%)
Year
Active Bankrupt
83

Figure 6.40: Average Beta
Finally, the correlations among the composites, return, volatility and beta were generally
weak (Table 6.28), suggesting that all these variables should be individually considered in
prediction.
Table 6.28: Correlation between Composite Indices and Stock Performance
Return Volatility Beta
Active Bankrupt Active Bankrupt Active Bankrupt
Return Active 1.00 0.73 0.37 0.75 0.03 -0.06
Bankrupt

1.00 0.33 0.53 0.17 -0.12
Volatility Active

1.00 0.61 0.24 -0.19
Bankrupt

1.00 0.25 0.25
Beta Active

1.00 0.16
Bankrupt

1.00
Market
Performance
(Composite Indices)
NYSE 0.48 0.19 -0.12 0.37 0.20 0.59
NASDAQ 0.49 0.13 0.00 0.33 -0.02 0.46
S&P 0.45 0.07 -0.04 0.33 0.11 0.61

6.8 Cross-Data Exploration
6.8.1 Correlation between Financial and Managerial Decision-Making Data
The relationship between financial and managerial decision-making data was investigated by
creating bar graphs for each managerial decision-making variable and the average value of
each financial variable. For example, Figure 6.41 correlates the possible categories of
related-party transactions (None or Yes) and average revenue. This relationship is
further broken down by state (i.e. active or bankrupt
12
). As illustrated, DMUs with related-
party transactions had greater revenue than those without, regardless of their state.

12
A DMU is classified bankrupt up to 3 years prior to filing Chapter 11.
-4.0
-2.0
0.0
2.0
4.0
6.0
8.0
1994 1996 1998 2000 2002 2004 2006 2008
Average
Beta
Year
Active Bankrupt
84

Over 150 graphs (25 financial 6 managerial decision-making variables) resembling
Figure 6.41 were generated to analyze the relationship between all financial and managerial
decision-making variables, and yielded similar conclusions (though some were less
statistically significant with high standard deviations). Results include:
Related-Party Transactions: Most variables
13
(i.e. COGS, SGA, II, IE, NI, Cash, Inv,
AR, CA, PPE, AD, TA, AP, CM, CL, LTD, TL, RE, SE and CFO) were similar to
revenue.
Legal Proceedings: Most variables were similar to revenue (Figure 6.42). Revenue
was higher without legal proceedings or with minor litigations compared to those
with significant proceedings, going concerns or bankruptcy filings, regardless of
state.
Auditors Opinion: Most variables were similar to revenue (Figure 6.43). Revenue
was higher with unqualified opinions compared to qualified opinions, regardless of
state.
Retirement Plan: Most variables were similar to revenue (Figure 6.44). Revenue
was higher for those with either a company-sponsored pension program or 401(k)
plan compared to those with none, regardless of state.
Auditor Change: Because a change of auditors was recorded from one year to the
next, the comparison of this managerial decision-making variable to all financial
items must be on a percentage change basis. Figure 6.45 correlates auditor change to
average percentage change in revenue over one year, and shows an increase in
revenue when there is a change. The difference was more pronounced between active
and bankrupt DMUs. Again, most variables were similar to revenue.
Turnover of Management: Figure 6.46 illustrates that those with management
turnover had less revenue, regardless of state. Also, most variables were similar to
revenue.

13
See List of Abbreviations for definitions of variable acronyms.
85

Figure 6.41: Relationship between Related Party
Transactions and Average Revenue

Figure 6.42: Relationship between Legal
Proceedings and Average Revenue

Figure 6.43: Relationship between Auditors Opinion
and Average Revenue

Figure 6.44: Relationship between Retirement Plan
and Average Revenue

Figure 6.45: Relationship between Auditor Change
and Average Change in Revenue

Figure 6.46: Relationship between Turnover of
Management and Average Change in Revenue

6.8.2 Correlation between Financial and Economic Data
Spearman correlation coefficients were calculated for each economic factor and the average
of each variable for a particular year by state (Table 6.29; and for E1-E17 codes, refer to
Table 6.24). The correlations for active and bankrupt DMUs were generally weak, with less
than 4.7% and 1.9% respectively, having coefficients greater than 0.6 (highlighted in shades
of green) or less than -0.6 (highlighted in shades of red).

0
500
1000
1500
2000
None Yes
$
(millions)
Active
Bankrupt
Average
0
500
1000
1500
2000
2500
$
(millions)
Active
Bankrupt
Average
0
500
1000
1500
2000
Unqualified Qualified
$
(millions)
Active
Bankrupt
Average
0
500
1000
1500
2000
2500
3000
None Company
Plan
401(k)
$
(millions)
Active
Bankrupt
Average
-10%
-5%
0%
5%
10%
15%
No Yes
%
Change
Active
Bankrupt
Average
0%
3%
6%
9%
12%
No Yes
%
Change
Active
Bankrupt
Average
86

Table 6.29: Correlation between Financial and Economic Variables

6.8.3 Correlation between Managerial Decision-Making and Economic Data
To determine the relationship between managerial decision-making data and economic
factors, the managerial decision-making variables were transformed for comparison. To do
so, the number of DMUs with active and bankrupt states for a given year for a given category
of a managerial decision-making variable was counted and then taken as a percentage with
Revenue 0.31 -0.40 -0.14 0.39 -0.71 0.52 0.40 0.01 0.29 -0.10 0.17 0.36 0.01 0.21 -0.06 -0.42 -0.23
Cost of Goods Sold -0.09 -0.18 -0.22 0.43 -0.58 0.18 -0.05 0.10 0.19 -0.07 0.11 0.09 0.02 -0.31 0.32 -0.40 -0.11
Selling, General and Administrative Expenses 0.51 -0.59 0.11 0.05 -0.65 0.57 0.64 -0.16 0.00 -0.03 0.06 0.19 -0.33 0.63 -0.44 0.02 -0.42
Interest Income 0.48 -0.15 0.27 0.22 -0.54 0.50 0.73 0.13 0.09 0.04 -0.06 0.09 -0.13 0.46 -0.19 0.05 -0.13
Interest Expense -0.01 0.40 0.04 -0.29 0.11 0.00 0.11 0.47 -0.25 0.00 -0.20 -0.48 -0.09 -0.17 0.51 0.25 -0.21
Income Tax Expense 0.03 -0.38 -0.36 -0.40 -0.02 0.03 0.05 -0.16 -0.27 0.00 0.05 -0.09 -0.40 0.34 -0.27 0.05 -0.31
Net Income 0.28 -0.35 -0.06 0.10 -0.23 0.36 0.24 -0.49 0.09 -0.26 0.28 0.31 -0.15 0.17 -0.15 -0.18 -0.04
Cash -0.04 0.14 -0.59 0.14 0.02 0.01 0.03 0.14 0.33 -0.35 0.21 0.19 0.19 -0.10 0.08 -0.55 0.35
Accounts Receivable 0.16 -0.03 0.18 -0.07 0.04 0.20 0.16 0.11 0.00 -0.04 -0.14 0.08 0.23 0.22 -0.11 0.10 -0.09
Inventories 0.49 -0.59 0.17 0.53 -0.85 0.69 0.53 -0.32 0.29 -0.15 0.28 0.44 -0.07 0.30 -0.25 -0.28 -0.20
Current Assets 0.52 -0.40 -0.43 -0.09 -0.36 0.51 0.63 0.03 0.07 0.04 0.01 0.41 -0.20 0.69 -0.40 -0.30 -0.22
Plant, Property and Equipment 0.16 -0.72 -0.05 0.26 -0.46 0.31 0.15 -0.70 -0.05 0.15 0.11 0.13 -0.36 0.25 -0.29 -0.41 -0.45
Accumulated Depreciation 0.32 -0.86 -0.31 0.04 -0.43 0.33 0.23 -0.58 -0.08 0.19 0.08 0.20 -0.50 0.49 -0.55 -0.49 -0.55
Total Assets 0.40 -0.73 -0.19 0.22 -0.72 0.56 0.44 -0.41 -0.10 0.16 0.02 0.29 -0.44 0.40 -0.28 -0.29 -0.41
Accounts Payable 0.54 -0.44 -0.01 0.08 -0.41 0.49 0.49 -0.12 0.26 -0.12 0.24 0.54 -0.09 0.52 -0.47 -0.22 -0.17
Current Maturities of Long-Term Debt 0.31 -0.20 -0.19 0.18 -0.41 0.40 0.40 0.11 0.48 -0.34 0.57 0.39 0.09 0.26 -0.18 -0.29 0.21
Current Liabilities 0.53 -0.43 -0.21 0.09 -0.43 0.58 0.56 -0.09 0.41 -0.38 0.49 0.58 0.00 0.53 -0.42 -0.35 0.00
Long-Term Debt 0.17 -0.04 0.66 0.17 0.01 0.14 -0.03 -0.33 -0.08 -0.03 -0.10 0.07 0.13 -0.10 -0.05 0.40 0.02
Total Liabilities 0.51 -0.53 -0.17 -0.10 -0.16 0.47 0.31 -0.32 0.12 -0.33 0.34 0.50 -0.17 0.47 -0.51 -0.23 -0.11
Retained Earnings -0.56 -0.01 -0.75 -0.34 0.40 -0.53 -0.44 -0.15 -0.15 0.11 0.00 -0.11 -0.29 -0.05 -0.01 -0.40 0.11
Shareholders' Equity 0.45 -0.53 -0.01 0.16 -0.76 0.58 0.64 -0.07 -0.12 0.21 -0.22 0.16 -0.34 0.51 -0.23 -0.08 -0.50
Cash Flow from Operating Activities 0.23 -0.50 -0.03 -0.27 -0.03 0.13 0.30 -0.44 -0.08 -0.13 0.17 -0.18 -0.38 0.58 -0.61 -0.08 -0.38
Cash Flow from Investing Activities 0.34 -0.53 -0.14 0.32 -0.43 0.53 0.38 -0.55 0.15 -0.10 0.30 0.46 -0.10 0.34 -0.29 -0.35 0.07
Cash Flow from Financing Activities 0.31 -0.04 -0.45 0.10 -0.05 0.30 0.39 0.05 0.58 -0.35 0.58 0.53 0.18 0.34 -0.23 -0.63 0.35
Change in Cash -0.40 0.18 -0.04 -0.10 0.19 -0.26 -0.26 0.14 0.09 0.00 0.13 -0.10 0.13 -0.13 0.22 -0.11 0.00
Revenue 0.17 -0.10 0.31 0.09 -0.57 0.33 0.45 0.08 -0.25 0.22 -0.30 -0.17 -0.23 0.16 0.19 0.30 -0.31
Cost of Goods Sold 0.07 0.04 0.31 0.10 -0.50 0.26 0.40 0.15 -0.28 0.28 -0.36 -0.22 -0.19 0.07 0.32 0.33 -0.22
Selling, General and Administrative Expenses 0.20 0.17 0.49 0.36 -0.45 0.44 0.46 0.13 0.10 -0.23 0.11 -0.11 0.20 -0.07 0.33 0.30 0.20
Interest Income -0.06 0.00 0.38 0.02 -0.17 -0.15 -0.01 0.13 -0.27 0.21 -0.23 -0.28 -0.28 -0.02 -0.11 0.47 0.01
Interest Expense 0.47 -0.18 -0.16 -0.41 -0.26 0.49 0.70 0.19 -0.43 0.16 -0.35 -0.05 -0.35 0.64 -0.07 0.21 -0.38
Income Tax Expense 0.12 -0.17 0.21 0.25 -0.30 0.15 0.12 -0.08 0.40 -0.49 0.55 -0.01 0.07 -0.03 -0.18 -0.13 0.07
Net Income 0.51 -0.12 -0.16 0.11 0.09 0.32 0.20 -0.09 0.31 -0.13 0.09 0.57 0.35 0.26 -0.32 -0.23 0.00
Cash 0.49 -0.38 -0.04 -0.20 -0.26 0.40 0.56 -0.06 -0.21 0.03 -0.09 -0.25 -0.37 0.54 -0.33 0.03 -0.51
Accounts Receivable 0.38 -0.02 0.24 -0.40 0.00 0.33 0.60 0.14 -0.24 -0.12 -0.05 -0.15 -0.27 0.60 -0.24 0.38 -0.12
Inventories 0.43 -0.32 0.08 0.17 -0.70 0.53 0.58 0.22 0.04 -0.01 -0.10 0.08 -0.09 0.36 -0.13 -0.02 -0.37
Current Assets 0.33 -0.37 0.16 0.09 -0.37 0.45 0.50 -0.16 0.09 -0.38 0.21 -0.05 -0.06 0.44 -0.36 0.00 -0.11
Plant, Property and Equipment 0.27 -0.40 -0.10 -0.04 -0.60 0.34 0.52 -0.03 -0.26 0.21 -0.27 -0.10 -0.45 0.44 -0.17 0.03 -0.43
Accumulated Depreciation 0.18 -0.52 -0.33 0.16 -0.54 0.21 0.15 -0.28 -0.17 0.13 -0.16 0.05 -0.43 0.09 -0.18 -0.21 -0.26
Total Assets -0.04 -0.15 0.38 -0.09 -0.31 0.10 0.13 0.03 -0.20 0.10 -0.20 -0.25 -0.07 0.11 0.05 0.34 -0.39
Accounts Payable -0.24 -0.13 -0.11 0.42 -0.37 0.06 -0.01 -0.02 0.04 -0.05 -0.02 -0.17 0.00 -0.15 0.16 -0.32 0.05
Current Maturities of Long-Term Debt -0.28 -0.17 0.19 0.07 -0.22 0.03 -0.13 -0.06 0.02 -0.38 0.21 -0.30 -0.04 -0.16 0.11 -0.08 -0.09
Current Liabilities -0.16 -0.05 0.26 0.32 -0.34 0.14 0.08 0.10 0.02 -0.12 -0.02 -0.20 0.11 -0.07 0.15 0.00 0.03
Long-Term Debt -0.51 -0.66 -0.32 -0.38 -0.16 -0.34 -0.34 -0.64 -0.35 0.05 0.04 -0.45 -0.59 0.25 -0.30 -0.19 -0.45
Total Liabilities -0.07 -0.22 0.46 -0.21 -0.22 0.00 0.05 -0.14 -0.42 0.25 -0.35 -0.41 -0.24 0.13 -0.01 0.52 -0.52
Retained Earnings 0.09 -0.37 -0.24 -0.12 -0.49 0.27 0.43 -0.11 -0.44 0.21 -0.29 -0.30 -0.65 0.35 0.00 -0.07 -0.39
Shareholders' Equity 0.29 -0.12 0.24 0.17 -0.49 0.52 0.60 0.02 0.10 -0.27 0.11 -0.02 0.08 0.28 0.02 0.10 -0.05
Cash Flow from Operating Activities 0.67 -0.06 0.13 0.15 -0.31 0.45 0.49 0.35 0.07 0.09 -0.17 0.24 0.12 0.30 -0.24 0.22 -0.07
Cash Flow from Investing Activities 0.02 -0.08 0.48 -0.12 -0.04 0.04 0.02 -0.14 0.18 -0.34 0.31 -0.10 0.17 0.04 -0.13 0.28 -0.16
Cash Flow from Financing Activities 0.38 -0.13 -0.29 -0.13 -0.17 0.17 0.38 0.02 -0.10 0.24 -0.03 -0.10 -0.34 0.25 -0.10 -0.03 -0.24
Change in Cash -0.31 0.11 0.19 0.15 -0.17 -0.18 -0.28 0.05 0.18 -0.12 0.18 -0.12 0.09 -0.44 0.33 -0.07 -0.05
Bankrupt
Active
87

respect of the total number of DMUs across all categories of the variable for the given year.
These percentages were then correlated to each economic factor (Table 6.30; and for E1-E17
codes, please to Table 6.24). The correlations between active and bankrupt DMUs were
generally weak, with less than 9.4% and 9.2% respectively, having coefficients greater than
0.6 (highlighted in shades of green) or less than -0.6 (highlighted in shades of red).
Table 6.30: Correlation between Managerial-Decision Making and Economic Variables

6.8.4 Correlation to Market and Stock Performance Data
Spearman correlation coefficients were computed for each financial variable and stock
performance measure (return, volatility and beta). In addition, the average of each financial
variable for a particular year was correlated to each composite index (NYSE, NASDAQ and
S&P) by state. Table 6.31 shows that the relationship between market and stock performance
and financial data is weak. In fact, the strongest correlations were between total liabilities of
bankrupt DMUs and the NYSE and S&P composite indices, as well as cash flow from
financing activities of active DMUs and all three composite indices.
None -0.05 -0.01 0.12 0.33 0.14 0.04 -0.05 -0.21 0.31 -0.40 0.19 0.22 0.35 0.03 -0.22 -0.25 0.33
Yes 0.05 0.01 -0.12 -0.33 -0.14 -0.04 0.05 0.21 -0.31 0.40 -0.19 -0.22 -0.35 -0.03 0.22 0.25 -0.33
Unqualified -0.60 0.03 -0.33 0.36 -0.02 -0.35 -0.63 -0.20 0.22 -0.13 0.32 0.00 0.14 -0.68 0.37 -0.44 0.35
Qualified 0.60 -0.03 0.33 -0.36 0.02 0.35 0.63 0.20 -0.22 0.13 -0.32 0.00 -0.14 0.68 -0.37 0.44 -0.35
None 0.19 -0.50 -0.52 0.19 -0.48 0.38 0.13 -0.04 0.32 -0.31 0.45 0.34 -0.01 0.11 -0.17 -0.57 -0.08
Insignificant -0.02 0.22 0.50 -0.46 0.33 -0.22 0.13 0.01 -0.61 0.47 -0.62 -0.47 -0.31 0.27 -0.06 0.76 -0.24
Significant -0.31 0.46 0.20 0.04 0.58 -0.45 -0.37 0.01 0.19 -0.19 0.04 -0.09 0.31 -0.28 0.02 -0.01 0.39
Going Concern -0.42 0.47 0.42 -0.12 0.13 -0.42 -0.38 0.21 -0.44 0.44 -0.44 -0.52 -0.12 -0.56 0.61 0.61 -0.01
None -0.18 0.30 -0.26 0.59 -0.24 0.12 -0.15 0.22 0.62 -0.42 0.55 0.18 0.51 -0.64 0.62 -0.61 0.45
Company Plan 0.10 -0.86 -0.19 -0.20 -0.31 0.20 0.07 -0.43 -0.37 0.06 -0.18 0.04 -0.52 0.54 -0.62 -0.09 -0.53
401(k) 0.02 0.38 0.52 -0.23 0.38 -0.24 0.02 0.04 -0.30 0.32 -0.43 -0.16 -0.01 0.07 -0.01 0.67 0.06
None -0.08 0.04 -0.07 0.57 -0.20 0.03 -0.17 0.07 0.63 -0.34 0.51 0.15 0.46 -0.38 0.16 -0.57 0.16
Yes 0.08 -0.04 0.07 -0.57 0.20 -0.03 0.17 -0.07 -0.63 0.34 -0.51 -0.15 -0.46 0.38 -0.16 0.57 -0.16
None -0.19 -0.54 -0.51 0.09 -0.11 0.04 -0.09 -0.69 0.01 -0.14 0.17 0.10 -0.23 0.15 -0.22 -0.61 -0.14
Yes 0.19 0.54 0.51 -0.09 0.11 -0.04 0.09 0.69 -0.01 0.14 -0.17 -0.10 0.23 -0.15 0.22 0.61 0.14
Yes 0.37 -0.08 -0.34 0.72 -0.70 0.57 0.31 0.26 0.48 -0.24 0.31 0.44 0.27 -0.24 0.28 -0.54 0.27
None -0.37 0.08 0.34 -0.72 0.70 -0.57 -0.31 -0.26 -0.48 0.24 -0.31 -0.44 -0.27 0.24 -0.28 0.54 -0.27
Unqualified . . . . . . . . . . . . . . . . .
Qualified . . . . . . . . . . . . . . . . .
None -0.18 -0.06 0.27 0.37 -0.42 0.02 -0.23 0.02 0.26 -0.07 0.26 0.03 0.25 -0.43 0.32 -0.04 -0.11
Insignificant -0.04 0.58 -0.22 0.63 0.05 0.04 -0.07 0.48 0.73 -0.34 0.37 0.49 0.73 -0.46 0.40 -0.55 0.76
Significant 0.08 -0.21 -0.12 -0.38 0.00 0.05 0.38 -0.19 -0.37 0.14 -0.32 -0.16 -0.42 0.59 -0.36 0.17 -0.17
Going Concern 0.23 -0.20 -0.25 -0.60 0.44 -0.03 0.07 -0.18 -0.27 0.13 -0.03 0.14 -0.31 0.51 -0.47 0.16 -0.13
Bankruptcy -0.03 -0.09 -0.04 -0.25 0.17 -0.16 -0.08 -0.12 -0.49 0.29 -0.35 -0.45 -0.46 0.03 -0.05 0.17 -0.26
None -0.04 0.58 -0.22 0.63 0.05 0.04 -0.07 0.48 0.73 -0.34 0.37 0.49 0.73 -0.46 0.40 -0.55 0.76
Company Plan 0.37 -0.12 -0.41 0.54 -0.61 0.52 0.25 0.31 0.57 -0.28 0.36 0.63 0.36 -0.13 0.15 -0.58 0.19
401(k) -0.19 -0.31 0.34 -0.77 0.33 -0.38 -0.12 -0.42 -0.85 0.47 -0.54 -0.66 -0.73 0.41 -0.39 0.72 -0.64
None 0.35 -0.13 -0.26 0.59 -0.56 0.46 0.29 0.24 0.45 -0.09 0.28 0.37 0.24 -0.07 0.13 -0.54 0.04
Yes -0.07 -0.20 -0.04 -0.50 0.38 -0.19 0.01 -0.46 -0.40 -0.02 -0.20 -0.09 -0.42 0.42 -0.49 0.28 0.03
None 0.26 -0.04 -0.68 0.08 0.00 0.25 0.26 0.13 0.45 -0.36 0.26 0.43 0.20 0.23 -0.17 -0.76 0.14
Yes -0.26 0.04 0.68 -0.08 0.00 -0.25 -0.26 -0.13 -0.45 0.36 -0.26 -0.43 -0.20 -0.23 0.17 0.76 -0.14
Active
Retirement Plan
Legal Proceedings
Change of Auditor
Related-Party
Transactions
Auditor's Opinion
Legal Proceedings
Change of Auditor
Retirement Plan
Management
Turnover
Management
Turnover
Related-Party
Transactions
Auditor's Opinion
Bankrupt
88

Table 6.31: Correlation between Market and Stock Performance and Financial Data

To determine the relationship between managerial decision-making data and market
and stock performance, the managerial decision-making variables were first transformed as
in Section 6.8.3. Generally, managerial decision-making variables were weakly correlated
with market and stock performance, with some exceptions highlighted in shades of red and
green (Table 6.33).
Finally, stock performance measures and composite indices were weakly correlated
with economic factors (Table 6.32; and for E1-E17 codes, please to Table 6.24), except for
volatility of bankrupt DMUs and GDP rate, as well as beta of bankrupt DMUs and cotton
price.

Return Volatility Beta Active Bankrupt Active Bankrupt Active Bankrupt
Revenue -0.08 -0.26 0.05 0.02 -0.35 0.15 -0.14 0.17 -0.28
Cost of Goods Sold -0.07 -0.25 0.02 -0.09 -0.31 0.10 -0.11 0.10 -0.24
Selling, General and Administrative Expenses -0.09 -0.25 0.07 -0.13 -0.08 0.03 -0.01 -0.11 -0.09
Interest Income -0.08 -0.25 -0.02 -0.02 -0.08 0.05 -0.21 0.00 -0.20
Interest Expense -0.04 -0.11 0.00 -0.19 -0.25 0.01 0.06 -0.15 -0.15
Income Tax Expense -0.13 -0.32 0.06 0.09 0.15 0.22 0.14 0.03 0.08
Net Income -0.12 -0.37 0.06 0.00 0.27 0.25 -0.02 0.08 0.21
Cash 0.02 -0.19 0.02 0.33 -0.09 0.46 -0.05 0.44 -0.17
Accounts Receivable -0.02 -0.04 0.05 -0.35 0.08 -0.40 0.09 -0.31 -0.03
Inventories -0.08 -0.30 0.00 -0.15 -0.24 0.02 -0.16 -0.02 -0.18
Current Assets -0.07 -0.32 0.02 0.19 -0.07 0.36 -0.05 0.29 -0.14
Plant, Property and Equipment -0.08 -0.31 0.01 -0.31 -0.25 -0.07 0.02 -0.21 -0.18
Accumulated Depreciation -0.08 -0.33 -0.02 -0.27 -0.28 -0.02 0.01 -0.17 -0.15
Total Assets -0.08 -0.30 0.04 -0.26 -0.57 0.09 -0.45 -0.09 -0.53
Accounts Payable -0.08 -0.25 0.02 0.03 -0.12 0.11 -0.04 0.12 -0.04
Current Maturities of Long-Term Debt -0.05 -0.22 0.00 0.45 -0.48 0.56 -0.25 0.48 -0.38
Current Liabilities -0.07 -0.23 0.03 0.29 -0.23 0.43 -0.25 0.37 -0.21
Long-Term Debt 0.00 -0.13 0.03 -0.46 -0.50 -0.56 0.09 -0.50 -0.41
Total Liabilities -0.07 -0.23 0.04 0.10 -0.71 0.18 -0.60 0.13 -0.72
Retained Earnings -0.10 -0.37 0.03 0.37 -0.27 0.54 0.17 0.42 -0.14
Shareholders' Equity -0.09 -0.34 0.04 -0.38 -0.16 -0.09 0.00 -0.22 -0.14
Cash Flow from Operating Activities -0.05 -0.33 0.04 -0.15 0.12 -0.02 -0.20 -0.26 0.00
Cash Flow from Investing Activities 0.11 0.22 -0.08 -0.04 -0.30 0.23 -0.27 0.09 -0.36
Cash Flow from Financing Activities -0.03 0.26 0.06 0.71 0.26 0.76 0.37 0.76 0.21
Change in Cash 0.13 -0.01 0.01 0.17 -0.26 0.12 -0.07 0.19 -0.14
NYSE NASDAQ S&P Stock Performance
89

Table 6.32: Correlation between Market and Stock Performance and Economic Factors

Table 6.33: Correlation between Market and Stock Performance and Managerial Decision-Making Data

6.9 Summary of Data Exploration
Results from data exploration showed that financial, managerial decision-making, stock,
market and economic data were individually poor predictors of bankruptcy and weakly
Active 0.20 -0.33 -0.28 -0.04 -0.18 0.26 0.28 -0.01 0.11 -0.30 0.35 0.21 -0.10 0.57 -0.34 -0.26 -0.03
Bankrupt -0.02 -0.10 0.33 -0.02 -0.13 0.02 0.06 0.05 0.15 -0.41 0.37 -0.06 0.07 0.27 -0.27 0.16 0.02
Active -0.34 -0.35 -0.15 0.15 -0.28 -0.08 -0.34 -0.19 -0.08 -0.06 0.17 -0.04 0.00 -0.04 0.12 -0.23 -0.23
Bankrupt -0.62 -0.12 -0.33 -0.04 0.15 -0.51 -0.53 -0.15 -0.13 0.14 0.17 -0.16 -0.08 0.14 -0.05 -0.12 0.02
Active -0.32 0.09 0.44 0.03 0.00 -0.19 -0.18 -0.06 -0.13 0.21 -0.02 -0.09 0.11 0.06 0.15 0.35 -0.01
Bankrupt -0.37 0.58 -0.36 0.17 0.38 -0.41 -0.22 0.22 0.38 -0.08 0.32 0.16 0.47 -0.20 0.27 -0.31 0.61
NYSE 0.11 0.27 -0.22 0.26 -0.07 -0.07 0.18 0.12 0.29 -0.28 0.33 0.39 0.24 0.18 0.01 -0.24 0.42
NASDAQ -0.02 0.16 -0.31 0.19 0.05 0.00 0.26 0.15 0.40 -0.34 0.43 0.29 0.25 0.02 0.20 -0.44 0.36
S&P 0.08 0.28 -0.31 0.31 -0.06 -0.02 0.22 0.22 0.36 -0.26 0.33 0.43 0.30 0.08 0.13 -0.41 0.47
Market Performance
(Composite Indices)
Return
Volatility
Beta
Return Volatility Beta Active Bankrupt Active Bankrupt Active Bankrupt
None 0.21 -0.08 0.13 0.25 0.32 0.09 0.21 0.28 0.16
Yes -0.21 0.08 -0.13 -0.25 -0.32 -0.09 -0.21 -0.28 -0.16
Unqualified -0.04 0.20 0.13 -0.13 -0.07 0.57 0.47 -0.16 0.29
Qualified 0.04 -0.20 -0.13 0.13 0.07 -0.57 -0.47 0.16 -0.29
None 0.03 0.26 0.18 0.45 0.15 0.60 0.24 -0.37 -0.13
Insignificant -0.20 -0.35 -0.38 -0.34 -0.08 -0.61 -0.29 0.36 -0.07
Significant 0.22 -0.06 0.11 0.01 0.22 -0.24 0.20 -0.01 0.39
Going Concern -0.47 -0.33 -0.42 -0.61 -0.31 -0.06 -0.39 0.03 -0.15
None 0.22 0.41 0.40 0.10 -0.03 0.49 0.31 -0.21 0.30
Company Plan -0.45 -0.36 -0.41 0.31 0.19 0.40 0.06 0.01 -0.67
401(k) 0.03 -0.23 -0.15 -0.50 -0.15 -0.71 -0.47 0.38 0.06
None 0.28 0.13 0.29 0.13 0.20 0.26 0.32 -0.09 0.21
Yes -0.28 -0.13 -0.29 -0.13 -0.20 -0.26 -0.32 0.09 -0.21
None -0.06 0.24 0.06 -0.03 -0.36 0.27 -0.01 0.06 -0.09
Yes 0.06 -0.24 -0.06 0.03 0.36 -0.27 0.01 -0.06 0.09
Yes 0.25 0.34 0.40 0.22 -0.09 0.22 0.04 -0.52 -0.02
None -0.25 -0.34 -0.40 -0.22 0.09 -0.22 -0.04 0.52 0.02
Unqualified . . . . . . . . .
Qualified . . . . . . . . .
None -0.26 -0.19 -0.17 -0.40 0.03 0.41 -0.31 0.28 -0.24
Insignificant 0.69 0.46 0.74 0.27 0.06 0.05 0.40 -0.06 0.65
Significant -0.11 0.08 -0.13 -0.01 0.12 -0.29 -0.39 -0.01 -0.02
Going Concern 0.24 0.18 0.14 0.29 -0.21 -0.27 0.39 0.08 0.08
Bankruptcy -0.20 -0.16 -0.25 0.15 -0.13 -0.24 0.31 -0.45 -0.19
None 0.69 0.46 0.74 0.27 0.06 0.05 0.40 -0.06 0.65
Company Plan 0.19 0.29 0.38 0.14 -0.08 0.41 -0.11 -0.38 -0.04
401(k) -0.57 -0.51 -0.71 -0.25 0.01 -0.29 -0.21 0.19 -0.42
None 0.41 0.30 0.46 0.29 -0.14 0.03 0.18 -0.14 0.02
Yes -0.13 0.00 -0.19 0.01 0.14 -0.28 -0.18 -0.09 -0.02
None 0.35 0.45 0.47 0.40 -0.08 0.16 0.12 -0.52 0.25
Yes -0.35 -0.45 -0.47 -0.40 0.08 -0.16 -0.12 0.52 -0.25
Stock Performance NYSE NASDAQ NYSE
Retirement
Plan
Legal
Proceedings
Change of
Auditor
Related-Party
Transactions
Auditor's
Opinion
Legal
Proceedings
Change of
Auditor
Retirement
Plan
Active
Bankrupt
Management
Turnover
Management
Turnover
Related-Party
Transactions
Auditor's
Opinion
90

correlated. This reinforced the motivation to combine the different data types as they may
add to and reflect different dimensions of company health.
91

7 Benchmarks
Based on a literature review, the bankruptcy prediction model developed in this thesis must
best Altmans model (Section 7.2) and that of the three major credit rating agencies (Section
7.3) because they are commonly used and well-regarded.

7.1 Definition of Type I and Type II Errors
To evaluate the effectiveness of a model, a confusion matrix (Table 7.1) can be generated
summarizing the count of true positive, true negative, false positive and false negative
instances. In this context, true positive was defined as predicting non-bankruptcy (active)
and being correct; true negative was defined as predicting bankruptcy and being correct; false
positive was defined as predicting non-bankruptcy (active) and being incorrect; and false
negative was defined as predicting bankrupt and being incorrect. A Type I Error is
equivalent to a false positive instance and a Type II Error is equivalent to a false negative
instance. From the confusion matrix, various evaluation measures can be computed
(Equations 7.1 to 7.5).
Table 7.1: Confusion Matrix
Actual
+
Predicted
+ True Positive (TP)
False Positive (FP)
Type I Error

False Negative (FN)
Type II Error
True Negative (TN)

True Positive Rate = Sensitivity =
TP
TP+FN
100%
(7.1)
False Positive Rate = Specificity =
FP
FP+TN
100%
(7.2)
Success Rate =
TP+TN
TP+FP+TN+FN
100%
(7.3)
Type I Error =
FP
FP+TN
100%
(7.4)
Type II Error =
FN
FN+TP
100%
(7.5)

92

7.2 Altmans Z-Score
Because Altmans bankruptcy model is widely used in industry (Section 2.3.1), it served as a
benchmark for this work. A Z-Score for each variation of Altmans model (manufacturing,
non-manufacturing, private) was computed for each of the 85 retail-apparel companies for
every year that data were available. Table 7.2 presents the average Z-Scores for all firms in
this sample from 1988 to 2009. While there was a distinction between average scores by
health, the high standard deviation indicated poor accuracy.
Table 7.2: Average Bankruptcy Scores of Retail-Apparel Companies by State
A DMU was classified bankrupt up to 3 years prior to filing of Chapter 11.
State Manufacturing (Original) Private Non-Manufacturing
Active 5.8 3.4 4.5 2.7 8.7 7.0
Bankrupt 2.7 2.1 2.8 1.3 1.6 4.0
All 5.6 3.4 4.4 2.7 8.2 7.0

Figure 7.1 shows the percentage of Type I errors by years prior to bankruptcy. For this
dataset on the retail-apparel industry, Altmans models did not perform well compared to his
original data. At its best, the non-manufacturing firms model yielded errors between 29-
80%, whereas the private firms and manufacturing firms (original) models, produced errors
of 79-100% and 50-87%, respectively. In addition, the Type II errors were higher (between
12-23%) than Altmans original study (Table 7.3). The discrepancies may be attributed to
Altmans models not being suited for the retail-apparel industry; the closest variation derived
for non-manufacturing firms and was not specific enough. Hence, there is a clear goal of
improving accuracy rates of 70%, 58% and 50%, when predicting cases today, and from one
and two years back, respectively.

Figure 7.1: Type I Error by Years Prior to Bankruptcy
0%
20%
40%
60%
80%
100%
0 1 2 3
Type I
Error
Years Prior to Bankruptcy
Private Firms Model with Retail-
Apparel Data
Public Manufacturing Firms Model
with Retail-Apparel Data
Public Non-Manufacturing Firms
Model with Retail-Apparel Data
Public Manufacturing Firms Model
with Altman's Original Data
93

Table 7.3: Type II Error
Model Type II Error
Private Firms with Retail-Apparel Data 14%
Public Manufacturing Firms with Retail-Apparel Data 23%
Public Non-Manufacturing Firms with Retail-Apparel Data 12%
Public Manufacturing Firms with Altmans Original Data 3-6%

7.3 Credit Rating Agencies
Three major credit rating agencies dominate 90-95% of the worlds credit rating market:
Moodys, Standard & Poors, and Fitch. These agencies issue specific bond credit ratings (in
three-letter codes) which assess the credit worthiness of a corporations or governments debt
challenges (Table D.2). A bond can also be broadly classified as either investment or non-
investment grade. The probability of corporate bond defaults is provided in Table 7.4. The
default rate of investment grade bonds is 4.1% and that of non-investment grade bonds is
42.4%; thus, Type I and II errors are 57.6% and 4.1%, respectively. These probabilities,
however, are based on all industries and not specific to retail-apparel.
Table 7.4: Cumulative Historic Corporate Default Rates (U.S. Municipal Bond Fairness Act, 2008)
Rating Categories Grade Moodys (%) Standard & Poors (%)
Aaa / AAA Prime 0.52 0.60
Aa / AA High 0.52 1.50
A / A Upper Medium 1.29 2.91
Baa / BBB Lower Medium 4.64 10.29
Ba / BB Non-Investment Speculative 19.12 9.93
B / B Highly Speculative 43.34 53.72
Caa C / CCC C
Substantial Risk, Extremely
Speculative or In Default
69.18 69.19
Investment Grade 2.09 4.14
Non-Investment Grade 31.37 42.35
All 9.70 12.98

94

8 Methodology
8.1 Application of Data Envelopment Analysis to Bankruptcy Prediction
Although traditionally used to measure productivity and efficiency of healthcare and
education systems, banks, manufacturing, benchmarking, franchises, management
evaluation, etc., DEA can be used to determine the state of each company in a sample within
a specific industry. To clarify DEA terminology in this thesis, a DMU (Decision Making
Unit) is associated with both a company and a year in the retail-apparel industry.
Combining a company and year is common DEA practice when analysis is across time
periods.
The power of DEA is its ability to act as a multicriteria sorting tool with the objective
function serving as a benchmark to organize and classify firms by their level of health. In
fact, for each DMU, DEA generates a score that quantifies a particular aspect of a companys
health, whether based on its financial operations or managerial performance, relative to that
of the sample and lies between 0 and 1. The difference in DEA scores is interpreted as
follows: a company X has a higher/lower risk of bankruptcy than company Y based on a
given set of criteria. The selection of variables is, therefore, of utmost importance because
inputs and outputs, and their definition and orientation, directly affect the frontier (i.e. the
benchmark) from which the value and magnitude of each score are determined.

8.1.1 Frontiers, Inputs and Outputs
The meaning of a DEA score depends on the composition of the frontier, either of DMUs
that are the best or worst performers within the sample.
Traditionally, a frontier comprises DMUs identified by DEA as best performers or
having the least risk of bankruptcy relative to the sample, and to these, DEA assigns a score
of 1. All other DMUs are scored as a fraction of one, with smaller values (corresponding to
farther distances from the frontier) indicative of poorer health and a higher risk of bankruptcy
(Figure 8.1). Outputs are defined as variables that improve a companys performance and are
maximized in the mathematical optimization, whereas inputs are defined as those that
increase its risk of bankruptcy and are minimized. Inputs and outputs are not necessarily
related (i.e. a decrease in liabilities does not translate to increase revenue) and generally,
these variables reflect on criteria that measure success and profitability.
95

Although less common, a frontier can consist of firms that face the greatest risk of
bankruptcy. Consequently, these poorest performing companies have a score of 1 while all
others are considered healthier and are scored as fractions of the worst companies. Inputs
and outputs have the reverse definitions, changing the objective to minimize outputs (now
defined as variables that increase the risk of bankruptcy) and maximize inputs (now defined
as those that decrease the risk of bankruptcy). There are two philosophies for selecting
variables for a worst performance frontier: inverse DEA (Figure 8.2) or worst-practice DEA.
Inverse DEA simply swaps the inputs and outputs of a normal DEA model whose variables
evaluate success and profitability. In contrast, Worst-Practice DEA, developed by Paradi et
al. (2004), selects a different set of variables that focus on measuring failure or insolvency.

Figure 8.1: DEA Frontier of Best Performers

Figure 8.2: DEA Frontier of Worst Performers

8.1.2 A New Classification Approach: A Modified Layering Technique
While DEA generates a score between 0 and 1 for every DMU, it is difficult to determine
whether there is a significant difference between scores that are relatively close to one
another (i.e. 0.80 vs. 0.81). As DMUs on the frontier have the same score of 1, Divine
(1986), Thanassoulis (1999) and Paradi et al. (2004) have used a layering technique where
these DMUs are removed or peeled from the sample and the analysis is re-run (Figure 8.3).
In normal DEA, each peel exposes a set of next best performing DMUs that form a new
frontier. Similarly, in inverse or worst-practice DEA, sequential layers of poor performance
are created with decreasing bankruptcy risk. In the three studies mentioned above, only 2-3
peels were made. This thesis was the first to create layers and peel DMUs off the frontier
until the following condition was no longer met: n maxm s, 3(m + s) where n, m and s
Input
(Cost of Goods Sold, etc.)
Output
(Revenue, etc.)
Frontier
Best Performers:
Lowest Risk of
Bankruptcy
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Low- Moderate
Risk
Moderate-High
Risk
Highest Risk of
Bankruptcy
Input
(Revenue, etc.)
Output
Frontier
Worst Performers:
Highest Risk of
Bankruptcy
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
*
Moderate-High
Risk
Low-Moderate
Risk
Lowest Risk of
Bankruptcy
96

are the number of DMUs, inputs and outputs, respectively (Banker et al., 1986). With this
process, we could conclude with certainty that one set of DMUs was better or worse than
another.

Figure 8.3: Layering Technique for Normal DEA

8.2 Overview: Metrics and the Second Stage Model
A metric was defined as a function of inputs and outputs, and was represented in the form of
a score generated by DEA. Nearly 60 combinations of inputs and outputs (V) derived from
annual reports for each of the 85 companies studied were applied to 60 SBM models which
generated 60 scores for each DMU (a firm in a given year) (Figure 8.4).

Figure 8.4: Example of Metrics and Second Stage Analysis
Input
Output
(Revenue, etc.)
*
*
*
* *
*
*
*
Layer 2 Frontier:
Second Best Performers,
Low Risk of Bankruptcy
Layer 3 Frontier:
Third Best Performers,
Moderate Risk of Bankruptcy
Remaining Un-layered:
Worst Performers,
Very High Risk of Bankruptcy
*
*
*
*
*
*
*
*
Layer 1 Frontier:
Best Performers,
No Risk of Bankruptcy
Layer 4 Frontier:
Fourth Best Performers,
High Risk of Bankruptcy
Metric B
Income Statement
SBM DEA Model
Balance Sheet
SBM DEA Model
Cash Flow Statement
SBM DEA Model
Managerial Decision-
Making
SBM DEA Model
Metric I
Metric C
Metric M
Each model creates a score or metric
corresponding to a given financial
statement and Notes, MD&A and
Auditors Report for each DMU
(company and year) in the sample.
Each firm has a set
of variables (V) for
each year data is
available used as
inputs and outputs
in different
combinations into
DEA models.
V1
V2
V3
V9
V10
V4
V5
V6
V7
V8
Overall
Health
The second-stage model
generates an overall
health score for each
company for a given year.
Second
Stage
SBM
DEA
Model
Metrics are outputs in a
second-stage DEA model
with a dummy input
variable.
97

Each score (metric) quantified a particular aspect of a firms health, relative to that of the
sample and was between 0 and 1. As an example, metric B evaluated a companys balance
sheet in a given year. Variables included inventories, total assets and total liabilities (as
inputs), retained earnings and/or shareholders equity (as outputs). Similarly, metric M
evaluated a firms managerial decision-making based on management turnover and/or legal
proceedings (as inputs).
While each metric can be considered independently, the combination of metrics into
one model in a second stage analysis
14
offered a more holistic view of health. Ultimately, 4
of the 60 metrics were combined by one model to generate an overall health score between 0
and 1 for each company. Specifically, the second stage model treated the 4 metrics as
outputs with a dummy input
15
, and created a frontier composed of the healthiest DMUs
(Chapters 10 and 12).

8.3 Variable Selection and Metrics Tested
The selection of variables is of utmost importance because inputs and outputs, and their
definition and orientation, directly affect the frontier from which the value and magnitude of
each score are determined. Financial variables considered in this thesis were inspired by past
studies that mainly aimed to measure profitability or insolvency as proxies for success or
failure, respectively (Table 8.1 and Table 8.2).
Another technique for choosing variables was principal component analysis (PCA)
which reduced over 50 financial variables to a set of 13, contributing to 5 components,
explaining 78% of the total variance (or dataset). The financial dataset was also reduced by
starting with all variables and removing one by one those that did not improve prediction
using logistic regression (LR). This subset of 14 variables slightly outperformed the PCA
group likely because of PCAs assumption of linearity.

14
The concept of a two-stage DEA model was applied by Paradi et al. (2011) for bank branch efficiency. He computed
productivity, profitability and intermediation scores of bank branches with 3 separate DEA models, then introduced them to
a second DEA calculation which generated an overall performance score that revealed areas for overall performance
improvement, and the tradeoffs among the 3 performance measures. A two-stage DEA model has also been used to assess
financial risk tolerance to capture its multidimensionality (Tran, 2007). Three elements of risk (attitude, capacity and
propensity) were measured independently then combined for overall risk tolerance. This worked well as no knowledge of
the relationship between the elements was required nor was there emphasis placed on any particular one.
15
A dummy variable is one where each DMU is given a value of 1 for that variable.
98

Table 8.1: Common Variables Measuring Good Performance or Success
Return on Sales
Operating Income
Revenue

Cash Return on Assets
Total Assets

Cash Flow Margin
Net Sales

Gross Margin
Revenue-Cost of Goods Sold
Revenue

Return on Common Equity
Net Income

Profitability
Return on Assets
Net Income + Interest 1 tax rate

Profit Margin
Net Income + Interest 1 tax rate
Sales

Efficiency
Total Assets Turnover
Sales

Accounts Receivable Turnover
Sales
Average Accounts Receivable

Inventory Turnover
Cost of Goods Sold
Average Inventory

Table 8.2: Common Variables Measuring Poor Performance or Failure
Liquidity
Current Ratio
Current Assets
Current Liabilities

Operating Cash Flow to
Current Liabilities
Average Current Liabilities

Solvency
Interest Coverage
Earnings Before Interest and Tax
Interest Expense

Liabilities to Assets
Total Liabilities
Total Assets

Liabilities to
Shareholders Equity
Total Liabilities

Cash Coverage
Operating Cash Flow Before Interest Tax
Interest Paid

Altmans
Z-Score
1.2
Net Working Capital
Total Assets
+1.4
Retained Earnings
Total Assets
+3.3
Earnings Before Interest and Taxes
Total Assets
+0.6
Market Value of Equity
Book Value of Total Debt
+
Sales
Total Assets

Quality of
Income
Net Income

In short, many different sets of variables were applied to SBM models. Those in
Table 8.3 are discussed in detail, and characterized by theme and frontier type. For a
complete list, see Appendix E.
99

Table 8.3: Abbreviated List of Metrics Tested
Theme Frontier Inputs Outputs
PCA Normal
AD, AR,CA, CL, COGS, Inv,
Other NonCA, PPE, SGA, TA, TL,
Rev, SE
LR Normal AP, CM, IE, PPE, SGA, TA, TL Cash, NI, Rev, SE
Profitability or Success
Normal
IE, TA CFO, NI, RE, Rev
CFO, NI, RE, Rev IE, TA
Inverse
COGS, IE, Inv, PPE, TA Rev, SE
Rev, SE COGS, IE, Inv, PPE,TA
Bankruptcy or Failure Worst-Practice CFO, EBIT IE
Income Statement Normal COGS, ITE, Net IE, SGA NI, Rev
Balance Sheet
Normal
AR, CA, Goodwill, Inv, MS, Net
PPE, LTIS, TA
Cash, RE, SE
Normal AP, CL, CM, LTD, NPSTD, TL RE, SE
Cash Flow Statement Normal Positive CFO, CFI; Negative CFF Negative CFO, CFI; Positive CFF
Making
Normal
Auditors Opinion, Auditor
Turnover, Legal Proceedings,
Management Turnover, Related
Parties, Retirement Plans
Dummy
16

8.4 Slacks-Based Measure of Efficiency (SBM) Model
The DEA model used in this work was the non-oriented Slacks-based measure of efficiency
(SBM) model with variable returns-to-scale. Unlike radial models, SBM captures the effect
of slacks (all inefficiencies) in the scores it produces. Each SBM efficiency score ] 1 , 0 [ is
a ratio of the average relative input consumption to the average relative output production
(Tone, 2001). Because SBM cannot handle zeros, zeros were replaced with a small positive
number. Also, variables with negative data (i.e. shareholders equity, net income) were
problematic because SBM is not translation invariant; that is, data cannot be scaled by adding
the absolute value of the largest negative number to remove all negatives. In these cases, for
any variable with negative data, two new variables (labelled positive and negative with
opposite orientations) were introduced in its place (Figure 8.5).
Net Income Positive Net Income (Output) Negative Net Income (Input)
$2,000,000 $2,000,000 $0
-$1,000,000 $0 $1,000,000
Figure 8.5: Example of Handling Negative Data
For instance, in normal DEA, a high net income was desirable. If net income was positive,
then its value was copied directly to the new variable positive net income and treated as an
output while the other new variable negative net income was assigned a zero and treated as
an input. In contrast, if net income was negative, then its absolute value was copied to

16
100

negative net income and treated as an input while positive net income was assigned a
zero and treated as an output.

8.5 Classification by Zones and Layering
The computer software, DEA-Solver-Pro Version 6.0, used to run DEA models was
purchased from SAITECH, Inc. It is a Microsoft Excel macro designed on the basis of the
textbook Data Envelopment Analysis A Comprehensive Text with Models, Applications,
References and DEA-Solver Software written by W.W. Cooper, L.M. Seiford and K. Tone
in 2000. Hence, input data files were in .xls form. Results generated by DEA-Solver-Pro
were also presented .xls form. SBM-NonOriented(SBM-V) was selected for all models.
DEA Solver Pro offers a lot of information on results (projections, weights), but this thesis
mainly focused on scores which were used to classify DMUs by zones or by layering.
When classifying by zones, DMUs were characterized as active or bankrupt based on
where their scores were relative to one or two cut-off values (Figure 8.6 and Figure 8.7,
respectively), which were selected to minimize Type I and II errors. The cut-off values
create zones similar to that of Altmans Z-Score: a DMU in the safe zone has little risk of
bankruptcy; a DMU in the danger zone has a high risk of bankruptcy; and a DMU in the grey
zone has a moderate risk of bankruptcy. From the cut-off values and zones, the numbers of
true positive, true negative, false positive and false negative instances were counted, and TP
rate, FP rate, success rate, Type I and II errors were calculated. A histogram of scores was
also created, and the averages and variances of all scores were computed by state.

Figure 8.6: Classification with One Cut-Off Value (x)

DEA Score
x
Subset of
Active
DMUs
Subset of
Bankrupt
DMUs
Type I Error
(False Positive)
True
Positive
True
Negative
Type II Error
(False Negative)
1 0
Safe
Zone
Danger
Zone
101

Figure 8.7: Classification with Two Cut-Off Values (x and y)
When classifying by layering (Section 8.1.2), true positive, true negative, false
positive and false negative instances were solely based on whether a DMU was on or away
from the frontier. In normal DEA, only a score of 1 represented a prediction of active; a
DMU with a score less than 1 remained unclassified until the next peel or run. The
evaluations (i.e. TP rate, FP rate, Type I and II errors, success rate) for a particular layer were
then calculated from the cumulative counts of the above instances, up to and including that
layer. For example, first layer evaluations were based only on the instances counted from the
first run, whereas second layer evaluations summed the instances counted from the first and
second peel. The third layer evaluations added the instances of the first, second and third
peel, and so on.

DEA Score
x
Subset of
Active
DMUs
Subset of
Bankrupt
DMUs
Type I Error
(False Positive)
y
Grey
Zone
True
Positive
True
Negative
Type II Error
(False Negative)
1 0
Safe
Zone
Danger
Zone
102

9 Preliminary Models: Results and Discussion
9.1 Principal Component Analysis and Logistic Regression Metrics
The first model applied financial variables selected by principal component analysis (PCA)
only (Table 8.3). Figure 9.1 presents the distribution of DEA scores based on a normal (best-
performing) frontier. Of the 701 DMUs
17
, 180 (or 26%) were identified as healthiest relative
to the sample and consequently given a score of 1; 4 of these DMUs, however, were actually
bankrupt or near failure
18
(Type I error). Also, the average scores were higher for active
DMUs and decreased as bankruptcy was imminent (Table 9.1).

Figure 9.1: Distribution of Scores Generated from PCA Normal Model
Table 9.1: Normal Scores Calculated with Variables Derived from PCA by State
State
19
Count Average Median Minimum Maximum
All A 651 0.70 0.23 0.68 0 1.00
All B 50 0.44 0.29 0.44 0 1.00
B 13 0.26 0.28 0.19 0 0.93
B-1 12 0.38 0.21 0.37 0 0.73
B-2 13 0.54 0.27 0.56 0 1.00
B-3 12 0.59 0.28 0.56 0.17 1.00
Total 701 0.69 0.25 0.67 0 1.00

With a kurtosis of -0.23 (i.e. a relatively flat distribution) and a skewness of -0.39 (i.e. a
distribution with an asymmetric tail extending toward negative values), two cut-off values (or
three zones) were used to classify financial health. Here, a DMU with a score was

17
There are 701 DMUs between 1996-2009 (Section 6.2).
18
A DMU was predicted bankrupt up to 3 years prior to filing Chapter 11.
19
Notation for possible DMU states is presented in Table 6.3.
0
25
50
75
100
125
150
175
200
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
1 Year Prior to Bankruptcy
2 Years Prior to Bankruptcy
Active
103

predicted bankrupt if 0 < 0.4 or active if 0.7 1. If 0.4 < 0.7, then it was
unclassified. Results from prediction by zones are summarized in Table 9.2. Type I and II
errors were computed to be 33% and 7%, respectively. Of the 43 active DMUs that were
misclassified as bankrupt, 14 of them represented years of firms that inevitably (after 3 or
more years) filed for Chapter 11; thus, the Type II error calculated was conservative. Also,
despite a success rate of 91%, 295 DMUs were unclassified a number that must be reduced
for this model to be effective.
Table 9.2: Confusion Matrix for PCA Normal Model

Actual

Active Bankrupt
Predicted
Active 355 8
Bankrupt 27 16

The PCA inputs and outputs were then swapped and applied to an inverse model for a
frontier of worst-performers. One hundred and six DMUs (15%) were identified as having
the highest risk of bankruptcy relative to the sample and were given a score of 1; 89 of these
DMUs, however, were actually active (Type II error). The average scores were lower for
active DMUs and increased as bankruptcy was imminent (Table 9.3). Unlike the normal
model, the distribution was bimodal (Figure 9.2). Hence, only one cut-off value separating
two zones was used for characterization: a DMU with a score was predicted as active if 0
< 0.7 or bankrupt if 0.7 1. All DMUs were classified (Table 9.4), yielding Type I and
II errors of 66% and 14%, respectively, with a success rate of 82%.
104

Figure 9.2: Distribution of Scores Generated from PCA Inverse Model
Table 9.3: Inverse Scores Calculated with Variables Derived from PCA by State
State Count Average Median Minimum Maximum
All A 650 0.14 0.34 0.002 0 1
All B 46 0.35 0.47 0.004 0 1
B 13 0.47 0.51 0.087 0 1
B-1 13 0.50 0.52 0.502 0 1
B-2 14 0.24 0.43 0.004 0 1
B-3 12 0.17 0.39 0.004 0 1
Total 700 0.16 0.36 0.002 0 1

Table 9.4: Confusion Matrix for PCA Inverse Model

Actual

Active Bankrupt
Predicted
Active 560 33
Bankrupt 90 17

The second model applied variables selected by logistic regression (LR) (Table 8.3).
The distribution of DEA scores based on a normal frontier (Figure 9.3) shows that of the 699
DMUs, 133 (or 19%) were identified as healthiest and given a score of 1; 4 of these DMUs,
however, were actually bankrupt or near failure
20
(Type I error) with 2 being identical to
those misclassified by the PCA model. With a kurtosis of -0.84 and skewness of 0.80, the
same two cut-off values (0.4 and 0.7) were used for classification, yielding Type I and II
errors of 11% and 71%, respectively.

20
0
100
200
300
400
500
600
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
105

Figure 9.3: Distribution of Scores Generated from LR Normal Model
When LR inputs and outputs were swapped and applied to an inverse model, the
results were similar to the PCA inverse model (Appendix F.1). With a bimodal distribution,
a cut-off value of 0.7 was used which produced Type I and II errors of 58% and 16%,
respectively.
Four conclusions came from the PCA and LR models (Table 9.5):
1. Inverse models created bimodal distributions which required only one cut-off value (2
zones) for classification. Normal models required at least 2 cut-off values (3 zones) to
minimize classification errors but were also conservative, leaving many DMUs
unclassified.
2. Type II error was generally lower than Type I error (exception: LR normal model). This
must be improved as Type I error is more costly.
3. The overall success rates were high, but this is deceiving because the sample had a higher
proportion of active DMUs which drove up the TP rate. In other words, the
misclassification of each bankrupt DMU was experienced more strongly.
4. There was nearly no relationship between the normal and inverse scores (Table 9.6) when
it was anticipated that the correlation would be -1. This unexpected result is explained in
Section 9.2.

0
20
40
60
80
100
120
140
160
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
106

Table 9.5: Summary of Results for Models with Variables Selected by PCA and LR

Evaluation
PCA Logistic Regression
Normal Inverse Normal Inverse
True Positive Rate 92.9% 86.2% 28.8% 84.2%
False Positive Rate 33.3% 66.0% 10.9% 58.0%
Type I Error 33.3% 66.0% 10.9% 58.0%
Type II Error 7.1% 13.8% 71.2% 15.8%
Success Rate 91.4% 82.4% 32.8% 81.2%
Unclassified 42.1% 0% 15.4% 0%

Table 9.6: Correlation between PCA and LR Scores

PCA Normal PCA Inverse LR Normal LR Inverse
PCA Normal 1 -0.15 0.66 -0.15
PCA Inverse

1 -0.18 0.72
LR Normal

1 -0.38
LR Inverse

1

9.2 Profitability Metrics
Over 30 combinations of financial variables were applied to normal and inverse DEA (for a
complete list, see Appendix E). Because many varied by one or two variables at a time, only
the best two models, denoted as A and B, are discussed. Figure 9.4 and Figure 9.5
summarize their results based on a normal frontier reflecting profitability. Again, a DMU
with a score was predicted bankrupt if 0 < 0.4 or active if 0.7 1. If 0.4 < 0.7,
then it remained unclassified. As expected, results changed with inputs and outputs. For
model A, the inputs were IE and TA, and the outputs were CFO, NI, RE and Rev. It fared
well with zero Type I error; however, Type II error was 79% and the success rate was only
29%. Also, over a quarter of the DMUs were unclassified. For model B, the inputs were
COGS, IE, Inv, PPE and TA, and the outputs were Rev and SE. It had considerably fewer
Type II errors (22%) and a vastly improved success rate (78%). However, these
improvements came with an increase of Type I error to 20% and the doubling of unclassified
DMUs. The difference between A and B is also rooted in dimensionality: A had 2
inputs and 4 outputs while B had 5 inputs and 2 outputs.
107

Metric A (Normal DEA)

Outputs:
CFO, NI, RE, Rev
Inputs:
IE, TA
Kurtosis: 0.09
Skewness: 0.95

Evaluation %
TP Rate 21.1
FP Rate 0
Type I Error 0
Type II Error 78.9
Success Rate 28.7
Unclassified 27.9

Figure 9.4: Summary of Normal Profitability Model A

Metric B (Normal DEA)

Outputs:
Rev, SE
Inputs:
COGS, IE, Inv, PPE, TA
Kurtosis: 0.15
Skewness: -0.04

Evaluation %
TP Rate 78.0
FP Rate 20.0
Type I Error 20.0
Type II Error 22.0
Success Rate 78.2
Unclassified 52.2

Figure 9.5: Summary of Normal Profitability Model B

Figure 9.6 and Figure 9.7 present the results of the inverse models of A and B
having similar classification zones as their normal counterparts; that is, a DMU with a score
was predicted active if 0 < 0.4 or bankrupt if 0.7 1, or remained unclassified if 0.4
< 0.7. With bimodal distributions, A and B had high Type I error (~74-78%) and
relatively low Type II error (<10%). Their success rate was high (>86%) but misleading
given the higher ratio of active to bankrupt DMUs. Also, over 90% of DMUs had scores less
than 0.1, driving the mean and median down to 0.08 and 0.0002, respectively. Overall,
inverse models were poor at classification.
0
50
100
150
200
250
300
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0
25
50
75
100
125
150
175
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
108

Metric A (Inverse DEA)

Outputs:
CFO, NI, RE, Rev
Inputs:
IE, TA
Kurtosis: 7.98
Skewness: 3.14

Evaluation %
TP Rate 93.5
FP Rate 78.0
Type I Error 78.0
Type II Error 6.5
Success Rate 88.4
Unclassified 0

Figure 9.6: Summary of Inverse Profitability Model A

Metric B (Inverse DEA)

Outputs:
Rev, SE
Inputs:
COGS, IE, Inv, PPE, TA
Kurtosis: 5.43
Skewness: 2.72

Evaluation %
TP Rate 91.5
FP Rate 74.0
Type I Error 74.0
Type II Error 8.5
Success Rate 86.9
Unclassified 0

Figure 9.7: Summary of Inverse Profitability Model B

The correlations in Table 9.7 show that normal and inverse models did not have a
perfect negative relationship.
Table 9.7: Correlation between Normal and Inverse Profitability Models

Normal Models Inverse Models

A B A B
Normal
Models
A 1 0.69 -0.66 -0.47
B

1 -0.33 -0.34
Inverse
Models
A

1 0.58
B

1

This inconsistency is attributed to the dimensionality of the DEA programming. To
illustrate, consider the following two non-oriented SBM models. In Figure 9.8, both axes
represent two output variables normalized by a unit input. A normal frontier (green curve)
0
100
200
300
400
500
600
700
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0
100
200
300
400
500
600
700
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
109

would thus consist of DMUs with the greatest amount of outputs produced given a unit input,
so DMUs C, D and E are the best performers while all others are less efficient. When the
variables are swapped for a two inputs and a unit output system, the model aims to minimize
the output and maximize inputs, yielding an inverse frontier with DMUs F and G. This
example illustrates the ideal situation where classification is consistent between normal and
inverse models; that is, the best performing DMUs in a normal model are not the worst
performing DMUs in its corresponding inverse model.

Normal Inverse

DMU
Outputs Input DEA Scores
Y1 Y2 X
Max Y,
Min X
Max X,
Min Y
A 4 3 1 0.83 0.66
B 7 3 1 0.62 0.91
C 8 1 1 1 0.54
D 4 2 1 1 0.49
E 2 4 1 1 0.47
F 10 1 1 0.90 1
G 3 7 1 0.62 1
Figure 9.8: Example of Consistent Normal and Inverse Models

However, inconsistent classification between normal and inverse models exists. In Figure
9.9, the normal frontier of the two outputs and a unit input model identifies B, E, F and G as
the best DMUs. When the variables are swapped for a two inputs and a unit output model,
the inverse model identifies A, D and G as the worst DMUs. Thus, G is present on both best
and worst frontiers. This contradiction suggests that a DMU can appear efficient when it
produces the greatest amount of outputs (Y
1
=6, Y
2
=2) for one unit of input (X), but at the
same time, it can look inefficient when it requires the same unit input (X =1) to produce less
outputs compared to another DMU, like B (Y
1
=2, Y
2
=7). This may be due to self-
identification of a unique DMU (one without peers). Also, G appears to be at the ends of
each space, serving as a boundary for both normal and inverse frontiers.
A
B
C
D
E
F
G
0
2
4
6
8
0 2 4 6 8 10
110

Normal Inverse

DMU
Outputs Input DEA Scores
Y1 Y2 X
Max Y,
Min X
Max X,
Min Y
A 1 5 1 0.33 1
B 2 7 1 1 0.61
C 3 4 1 0.69 0.92
D 4 3 1 0.67 1
E 4 6 1 1 0.54
F 5 5 1 1 0.60
G 6 2 1 1 1
Figure 9.9: Example of Inconsistent Normal and Inverse Models

As the complexity and dimensionality of DEA programming increases with the
application of more variables, inconsistency becomes more common and further establishes
that normal and inverse models are not opposites. For the 2 inputs and 2 outputs model
presented in Table 9.8, the weak relationship between normal and inverse scores is clear as
A, B, G, L, N and P are identified as both best and worst performers. Inconsistencies do not
necessarily imply that these DMUs are outliers but rather, confirm that the selection of
variables and model/frontier greatly influence results and classification.
Table 9.8: Example of Inconsistent Normal and Inverse Models of a Two Inputs Two Outputs System
DMU
Inputs Outputs DEA Scores
X1 X2 Y1 Y2
Max Y, Min X
(Normal)
Max X, Min Y
(Inverse)
A 1 4 1 5 1 1
B 2 10 2 7 1 1
C 2 2 3 4 1 0.56
D 3 4 4 3 0.57 1
E 4 7 4 6 0.57 1
F 4 3 5 5 0.79 0.52
G 4 4 6 2 1 1
H 5 7 7 9 1 0.62
I 5 8 7 3 0.47 1
J 6 6 8 12 1 0.59
K 7 9 9 9 0.69 1
L 8 2 9 8 1 1
M 8 7 10 10 0.77 1
N 9 1 11 1 1 1
O 9 3 12 6 1 0.94
P 10 2 12 9 1 1

9.3 Combining Normal and Inverse Scores
The weak correlation between normal and inverse models alluded to a possible
complementary relationship and encouraged the combination of their scores to increase
success rate, decrease Type I and II errors, and reduce the number of unclassified DMUs.
A
B
C
D
E
F
G
0
2
4
6
8
0 2 4 6 8
111

Two novel approaches were created to combine normal and inverse scores generated from
A and B in Section 9.2: parallel classification and serial classification. Parallel
classification considered both normal and inverse scores in combination (Figure 9.10). A
DMU was predicted active if its normal score was between 0.7 and 1 and its inverse score
was between 0 and 0.4. It was predicted bankrupt if its normal score was between 0 and 0.4
and its inverse score was between 0.7 and 1. If neither of these conditions was satisfied, the
DMU was unclassified.

Figure 9.10: Parallel Classification
Table 9.9 summarizes the results of parallel classification of A and B. The performance
of the combination of scores (Normal Inverse) appeared to be an average of normal and
inverse scores. In fact, the strengths of a normal model (low Type I error, high success rate)
were compromised when combined with the inverse model. At the same time, its
weaknesses (high Type II error) were reduced. Similarly, the strengths of an inverse model
(low Type II error) were reduced when combined with a normal model, while its weaknesses
(high Type I error, low success rate) were improved. Furthermore, the percentage of
unclassified DMUs increased because of the inconsistency of classification between normal
and inverse models on their own.
Table 9.9: Parallel Classification of Models A and B
Evaluation
(%)
Model A Model B
Normal Inverse Normal Inverse Normal Inverse Normal Inverse
TP Rate 21.1 93.5 72.9 78.0 91.5 96.6
FP Rate 0 78.0 0 20.0 74.0 45.5
Type I Error 0 78.0 0 20.0 74.0 45.5
Type II Error 78.9 6.5 27.1 22.0 8.5 3.4
Success Rate 28.7 88.4 75.0 78.2 86.9 96.6
Unclassified 27.9 0 80.0 52.2 0 56.3

To reduce the occurrence of unclassified DMUs, two serial approaches were tested.
The first strategy involved predicting with normal scores as before (active if 0.7 1;
Normal
DEA
Inverse
DEA
Predict as Active
Predict as Bankrupt
Unclassified
0.7
norm
1 0
inv
< 0.4
0
norm
< 0.4 0.7
inv
1
All Other Conditions
112

bankrupt if 0 < 0.4; and unclassified if 0.4 < 0.7), then sorting any unclassified DMUs
by their inverse scores (Figure 9.11). For both A and B, serial classification (Normal
Inverse) outperformed normal models in all categories except Type I error (Table 9.10). This
is because the subsequent prediction (by inverse scores) of DMUs unclassified (according to
normal scores) tended towards active classification (n.b. recall that 90% of inverse scores
were less than 0.1).

Figure 9.11: Serial Classification (Normal then Inverse DEA)
Table 9.10: Serial Classification of Models A and B (Normal then Inverse DEA)
Evaluation
(%)
Model A Model B
Normal Normal Inverse Normal Normal Inverse
TP Rate 21.1 43.5 78.0 92.6
FP Rate 0 2.0 20.0 62.0
Type I Error 0 2.0 20.0 62.0
Type II Error 78.9 56.2 22.0 7.4
Success Rate 28.7 47.6 78.2 88.7
Unclassified 27.9 0 52.2 0

The second type of serial classification ordered DMUs by their inverse scores first
(bankrupt if 0.7 1; active if 0 < 0.4; and unclassified if 0.4 < 0.7), then sorted
those unclassified by their normal scores (Figure 9.12). For both A and B, serial
classification (Inverse Normal) performed identically to inverse models alone (Table
9.11). This is because all DMUs were already classified by the inverse model, making
subsequent prediction by normal scores obsolete.

0
norm
< 0.4
0.7
norm
1
0.4
norm
< 0.7
Normal
DEA
0
inv
0.4
0.7
inv
< 1
Predict as Active
Predict as Bankrupt
Unclassified
Inverse
DEA
0.4
inv
< 0.7
113

Figure 9.12: Serial Classification (Inverse then Normal DEA)
Table 9.11: Serial Classification of Models A and B (Inverse the Normal DEA)
Evaluation
(%)
Model A Model B
Inverse Inverse Normal Inverse Inverse Normal
TP Rate 93.5 93.5 91.5 91.5
FP Rate 78.0 78.0 74.0 74.0
Type I Error 78.0 78.0 74.0 74.0
Type II Error 6.5 6.5 8.5 8.5
Success Rate 88.4 88.4 86.9 86.9
Unclassified 0 0 0 0

9.4 Insolvency Metrics
In addition to creating metrics that reflect profitability, sets of variables commonly used to
measure insolvency were also tested. Because of their similarity, the results of one metric
only, denoted C are discussed (Figure 9.13). This worst-practice model had one output
(IE) and two inputs (CFO, EBIT). Like inverse models, a DMU with a score was predicted
active if 0 < 0.4 or bankrupt if 0.7 1. If 0.4 < 0.7, then it was unclassified.
Generally, worst-practice models were ineffective at prediction because nearly 95% of
DMUs had scores less than 0.1, making the separation of active and bankrupt DMUs
difficult. Consequently, Type I error was high. And while the success rate was high, it was
misleading given the high ratio of active to bankrupt DMUs.
Moreover, insolvency metric C was combined with profitability metrics A and
B using both parallel and serial classification techniques (Table 9.12, Table 9.13 and Table
9.14). Noting that C was similar to the inverses of A and B, when it was combined
with normal A and B, prediction did not improve. Hence, worst-practice DEA was not
applied in this thesis beyond this exercise.

0.7
inv
1
0
inv
< 0.4
0.4
inv
< 0.7
Inverse
DEA
0.7
norm
1
0
norm
< 0.4
Predict as Active
Predict as Bankrupt
Unclassified
Normal
DEA
0.4
norm
< 0.7
114

Metric C (Worst-
Practice DEA)

Outputs:
IE
Inputs:
CFO, EBIT
Kurtosis: 29.8
Skewness: 5.45

Evaluation %
TP Rate 98.1
FP Rate 91.5
Type I Error 91.5
Type II Error 1.9
Success Rate 92.1
Unclassified 0.9

Figure 9.13: Summary of Solvency Worst-Practice Model C

Table 9.12: Parallel Classification of Normal and Worst-Practice Models
Evaluation
(%)
Normal A Worst-Practice C Normal B Worst-Practice C
A C A C B C B C
TP Rate 21.1 98.1 92.2 78.0 98.1 98.4
FP Rate 0 91.5 0 20.0 91.5 75.0
Type I Error 0 91.5 0 20.0 91.5 75.0
Type II Error 78.9 1.9 7.8 22.0 1.9 1.6
Success Rate 28.7 92.1 92.5 78.2 92.1 96.5
Unclassified 27.9 0.9 84.7 52.2 0.9 54.6

Table 9.13: Serial Classification of Normal then Worst-Practice Models
Evaluation
(%)
Normal A Worst-Practice C Normal B Worst-Practice C
A A C B B C
TP Rate 21.1 44.1 78.0 95.4
FP Rate 0 2.0 20.0 68.1
Type I Error 0 2.0 20.0 68.1
Type II Error 78.9 55.9 22.0 4.6
Success Rate 28.7 47.9 78.2 91.1
Unclassified 27.9 0 52.2 0.9

Table 9.14: Serial Classification of Worst-Practice then Normal Models
Evaluation
(%)
Worst-Practice C Model A Worst-Practice C Model B
C C A C C B
TP Rate 98.1 97.7 98.1 98.1
FP Rate 91.5 86.0 91.5 91.5
Type I Error 91.5 86.0 91.5 91.5
Type II Error 1.9 2.3 1.9 1.9
Success Rate 92.1 91.7 92.1 92.1
Unclassified 0.9 0 0.9 0.9

0
100
200
300
400
500
600
700
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
115

9.5 Summary of Preliminary Models
The preliminary models of Sections 9.1 to 9.4 provided valuable insight that was essentially
referenced to develop more comprehensive and complex prediction models in this thesis.
The lessons learned were:
Normal models yielded lower Type I errors and higher overall success rates.
Inverse models produced lower Type II errors and fewer unclassified DMUs.
Normal and inverse models were not necessarily consistent.
When combining normal and inverse scores, the weaknesses of each model were
improved but at the expense of reducing their strengths. In other words, parallel and
serial classification simply averaged the performance of normal and inverse models.
Worst-practice models were similar in distribution to inverse models, having low Type II
errors but high Type I errors.
Ultimately, all modelling (from this point forward) was based on normal DEA only in order
to focus on lowering Type I errors, which are more costly than Type II errors.

116

10 Financial Statement and Managerial Decision-Making Models: Results
and Discussion
While most studies focus on measuring profitability and/or insolvency, a different approach
was taken in this thesis: creating individual metrics that reflect a particular aspect of an
annual report. All DEA models in this and subsequent chapters were SBM and the scores
generated were based on a normal frontier.

10.1 Income Statement Metric
The metric (IS) based on the income statement included 4 inputs (COGS, ITE, Net IE, SGA)
and 2 outputs (NI and Rev). Results from classification by zones
21
(Figure 10.1) show IS
performed similarly to profitability metric B because they shared common variables.
However, IS may be a better predictor: reducing Type I error (from 23% to 12%) and the
unclassified rate (from 47% to 21%) while increasing Type II error (from 20% to 32%).
Benchmarking against Altmans model, IS lowered Type I error by 85% but doubled Type II
error.

IS Metric (Normal DEA)

Outputs:
NI, Rev
Inputs:
COGS, ITE, Net IE, SGA
Kurtosis: -1.26
Skewness: -0.29

Evaluation %
TP Rate 68.2
FP Rate 11.9
Type I Error 11.9
Type II Error 31.8
Success Rate 69.7
Unclassified 21.3

Figure 10.1: Summary of Income Statement (IS) Metric

In addition, classification by layering was applied (Sections 8.1.2). For prediction from one
year back, Figure 10.2 shows how Type I and II errors, and active and bankruptcy

21
A DMU with a score was predicted bankrupt if 0 < 0.4 or active if 0.7 1, or remained unclassified if 0.4 <
0.7.
0
25
50
75
100
125
150
175
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
117

classifications
22
, changed with each subsequent layer. Recall that a goal was to create a
metric with a layer where both of its errors (red curves) were less than (below) Altmans
model (blue lines). The effectiveness of each model developed in this thesis was determined
by this measure.
There were trade-offs between Type I and II errors, and accurate classifications of
active and bankrupt DMUs. After the first run, bankruptcy classification and Type II error
were at their maximum whereas active classification and Type I error were at their minimum.
With each subsequent peel, bankruptcy prediction worsened (Type I error rose) and active
prediction improved (Type II error fell). The lines intersected after roughly 3 runs,
suggesting that at its best, IS predicted all DMUs 58% correctly. This error rate was identical
to that of Altmans model. Finally, when attempts at classification started from two and
three years back (Appendix F.2), the model was less accurate and errors increased.

Figure 10.2: Prediction by IS Metric from One Year Back
A DMU was predicted bankrupt up to a year prior to filing Chapter 11.

10.2 Balance Sheet Metrics
Because of the abundance of variables on the balance sheet, two metrics were created
reflecting assets and liabilities separately.
Figure 10.3 summarizes the results from classification by zones
23
reflecting assets
(BSA) with 8 inputs (AR, CA, Goodwill, Inv, MS, Net PPE, LTIS, TA) and 3 outputs (Cash,

22
Active classification = 100% Type II error; Bankruptcy classification = 100% Type I error
23
0.7.
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
Bankruptcy Classification
Active Classification
Altman: Type I Error
Altman: Type II Error
118

RE, SE). BSA performed similarly to profitability metric A, and compared to Altmans
model, it reduced Type I error by three times but with over six times more Type II error.
Furthermore, Figure 10.4 shows the effectiveness of the prediction by layering from one year
back (see Appendix F.3 for two and three years back). The intersections after 4 peels reveal
that BSA correctly classified 78%, 72% and 70% of all DMUs, when attempting prediction
one, two and three year(s) back, respectively all better than Altmans model.

BSA Metric (Normal DEA)

Outputs:
Cash, RE, SE
Inputs:
AR, CA, Goodwill, Inv,
MS, Net PPE, LTIS, TA
Kurtosis: -0.39
Skewness: 0.93

Evaluation %
TP Rate 22.3
FP Rate 6.7
Type I Error 6.7
Type II Error 77.7
Success Rate 27.7
Unclassified 21.0

Figure 10.3: Summary of Balance Sheet Assets (BSA) Metric

Figure 10.4: Prediction by BSA Metric from One Year Back
0
50
100
150
200
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8
Layer
Type I Error
Type II Error
119

Figure 10.5 summarizes the results from classification by zones
24
reflecting liabilities
(BSL) with 6 inputs (AP, CL, CM, LTD, NPSTD, TL) and 2 outputs (RE, SE). BSL
performed like profitability metric A, having zero Type I error. The effectiveness of
prediction by layering from one year back is presented in Figure 10.6 (see Appendix F.4 for
two and three years back). With 7 peels, BSL correctly classified 78%, 72% and 70% of all
DMUs when attempting prediction one, two and three year(s) back, respectively an
improvement over Altmans model.

BSL Metric (Normal DEA)

Outputs:
RE, SE
Inputs:
AP, CL, CM, LTD,
NPSTD, TL
Kurtosis: 0.83
Skewness: 1.20

Evaluation %
TP Rate 12.7
FP Rate 0
Type I Error 0
Type II Error 87.3
Success Rate 19.9
Unclassified 22.8

Figure 10.5: Summary of Balance Sheet Liabilities (BSL) Metric

Figure 10.6: Prediction by BSL Metric from One Year Back

24
0.7.
0
50
100
150
200
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Layer
Type I Error
Type II Error
120

10.3 Cash Flow Statement Metric
Three variables from the cash flow statement were considered: cash flow from operating
activities (CFO), cash flow from investing activities (CFI), and cash flow from financing
activities (CFF). Defining their orientation was a challenge because the sign (positive or
negative) of these variables typically indicate different stages of companys life cycle as
opposed to good or poor health (Figure 10.7). Since firms in this sample studied were
publicly traded, it was assumed that they were more mature. Thus, good health was
associated with positive CFO, positive CFI and negative CFF (and were outputs in DEA)
while poor health was linked to negative CFO, negative CFI and positive CFF (inputs).
However, after applying this orientation of variables to DEA, no reliable metric was
produced from the cash flow statement. In fact, the distribution of scores mimicked that of
inverse and worst-practice DEA models (Figure 9.6, Figure 9.7 and Figure 9.13).

Figure 10.7: Cash Flows in Product Life Cycle (Stickney et al., 2007)

10.4 Managerial Decision-Making Metric
Figure 10.8 presents the results from classification by zones
25
reflecting managerial decision-
making (MDM) with 6 inputs (auditors opinion, auditor turnover, legal proceedings,
management turnover, related parties and retirement plans, Table 6.19) and a unit output.
MDM had the lowest Type II error but a high Type I error. Figure 10.9 plots the results from
the prediction by layering from one year back (see Appendix F.5 for two and three years
back). With 2 peels, it correctly classified 65%, 63% and 61% of all DMUs when attempting
prediction one, two and three year(s) prior, respectively.

25
0.7.
Introduction Growth Maturity Decline
$
Cash Flow from
Financing Activities
Cash Flow from
Operating Activities
Cash Flow from
Investing Activities
121

MDM Metric (Normal DEA)

Outputs:
Dummy
Inputs:
Auditors Opinion, Auditor
Turnover, Legal
Proceedings, Management
Turnover, Related Parties,
Retirement Plans
Kurtosis: -0.05
Skewness: -0.58

Evaluation %
TP Rate 97.3
FP Rate 82.6
Type I Error 82.6
Type II Error 2.8
Success Rate 92.9
Unclassified 15.7

Figure 10.8: Summary of Managerial Decision-Making (MDM) Metric

Figure 10.9: Prediction by MDM Metric from One Year Back

10.5 Trade Data Metric
In addition to creating metrics with information collected from the annual report, stock
performance data were considered. Specifically, for each DMU, the annual return was
applied as an output while volatility and beta were inputs to DEA. However, no reliable
metric could be produced because the distribution of the scores was similar to inverse and
0
50
100
150
200
250
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
122

worst-practice DEA models (Figure 9.6, Figure 9.7 and Figure 9.13); that is, the majority of
scores were less than 0.1 and offered little room for discrimination.

10.6 Correlation between Metrics
The correlations between scores of metrics IS, BSA, BSL and MDM after the first run were
weak (Table 10.1), supporting the hypothesis that they reflect different aspects of company
health and should therefore be considered altogether for a holistic view of company health.
Table 10.1: Correlation between Metrics after First Run
All correlations were significant at 0.01 level.

Income
Statement
Balance
Sheet Assets
Balance Sheet
Liabilities
Managerial
Decision-Making
Income Statement 1 0.49 0.47 0.14
Balance Sheet Assets

1 0.71 0.08
Balance Sheet Liabilities

1 0.15
Managerial Decision-Making

1

10.7 The Effect of Inflation
As mentioned in Section 6.2, all studies in bankruptcy research compare companies across
different years without calibrating for the time value of money. Although creating a
technique that calibrates for time is beyond the scope of this thesis, the effect of inflation was
considered.
Financial statements are typically reported in actual dollars. To convert actual dollars
into real dollars relative to a base year, the following formula was applied:
R
0, N
=
A
N
I
0,N
/ 100

where A
N
represents actual dollars in year N;
R
0,N
represents the real dollars equivalent to A
N
relative to a base year 0; and,
I
0,N
is the value of a global price index at year N relative to a base year 0.
In this thesis, the Consumer Price Index with a base 1982 of 100. The values of all
the financial variables were transformed accordingly, and the three metrics reflecting the
financial statements were re-calculated. Figure 10.10, Figure 10.11 and Figure 10.12
compare IS, BSA and BSL scores based on actual (ignoring inflation) and real (adjusted for
inflation) dollars, after the first run. These non-layered results show that the effect of
inflation is insignificant as the correlations between results in actual and real dollars are
123

strong (0.96 to 1), and the evaluation measures differ by less than 5%. Furthermore,
layering results did not change when adjusted for inflation. For instance, the IS metric
maintained a prediction accuracy around 58-60% (Figure 10.2 without inflation while Figure
10.13 with inflation). The likely reason the time value of money being negligible is likely
due to the fact that the objective function in DEA is fractional, thereby dividing out the
effect of inflation.

Pearsons Correlation Coefficient
= 0.99 (sig. at the 0.01 level)
Evaluation
Actual
(%)
Real
(%)
TP Rate 68.2 68.3
FP Rate 11.9 14.3
Type I Error 11.9 14.3
Type II Error 31.8 31.7
Success Rate 69.7 69.7
Unclassified 21.3 20.5
Figure 10.10: The Effect of Inflation on IS Scores

Evaluation
Actual
(%)
Real
(%)
TP Rate 22.3 22.6
FP Rate 6.7 6.7
Type I Error 6.7 6.7
Figure 10.11: The Effect of Inflation on BSA Scores

Evaluation
Actual
(%)
Real
(%)
TP Rate 12.7 13.2
FP Rate 0 0
Type I Error 0 0
Figure 10.12: The Effect of Inflation on BSL Scores

y = 0.99x + 0.01
R = 0.98
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
IS Score
based on
Real Dollars
IS Score based on Actual Dollars
y = 0.96x + 0.01
R = 0.93
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
BSA Score
based on
Real Dollars
BSA based on Actual Dollars
y = 1.00x
R = 0.99
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
BSL Score
based on
Real Dollars
BSL Score based on Actual Dollars
124

Figure 10.13 Prediction by IS Metric from One Year Back and Based on Real Dollars

0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7
Layer
Type I Error
Type II Error
125

11 Market and Economic Factors: Results and Discussion
Determining the relationships between financial and managerial decision-making data, and
market and economic (ME) data are critical: if ME data is not inherent in financial and
managerial decision-making values, then they should be added as separate variables to
prediction models. Although 20 ME factors were already selected from the original list of
over 100, this set was further reduced because it outnumbered the financial and managerial
decision-making variables. Specifically, aggregate indicators were created representing
different aspects of the market and economy that could then be regressed with DEA scores
from IS, BSA, BSL and MDM models, or used in sensitivity analysis to examine how firms
react to non-controllable factors. The most common method to reduce the number of
variables is factor analysis; however, the number of DMUs (i.e. years) was insufficient. As a
result, a novel DEA approach was developed to create 5 new indicators reflecting the general
economy, apparel industry, commodity prices, housing market, and market performance. In
other words, 20 individual ME factors were used as inputs and outputs in SBM models with
scores generated reflecting 5 broader categories.

11.1 General Economic Indicator
To create an aggregate general economic indicator, 2 inputs (unemployment rate, debt as a
percentage of GDP) and 3 outputs (GDP growth rate, inflation, prime interest rate) were
applied to an SBM model
26
. Figure 11.1 and Figure 11.2 plot the values in percentage or rate
of factors on the primary y-axis. The factors are separated into two graphs because the range
of debt as a percentage of GDP is higher (50-90%) than that of others (0-10%). The score
generated for each year is an aggregate measure of economic performance for that given year
and is represented by a black line on the secondary y-axis.
The strong correlations to this aggregate economic indicator (Table 11.1) proved that
the reduction of five variables to one was appropriate.

26
Negative values of the variable were rounded to zero as opposed to separating the variable into its positive and negative
components because the latter unnecessarily creates a second variable (a negative component) with most entries equal to
zero.
126

Figure 11.1: General Economic Performance by Year I

Figure 11.2: General Economic Performance by Year II
Table 11.1: Correlations to Aggregate General Economic Indicator
Aggregate General Economic Indicator
GDP Growth Rate 0.69
Debt as Percentage of GDP -0.72
Inflation 0.68
Interest Rate 0.93
Unemployment Rate -0.62

11.2 Apparel Industry Indicator
To create an aggregate apparel industry indicator, 4 outputs (personal consumption
expenditures: clothing, GDP: clothing, apparel labour productivity, apparel imports) were
applied with a dummy input to an SBM model (Figure 11.3). Again, the correlation between
each factor and the aggregate apparel industry indicator (Table 11.2) showed that the
reduction was appropriate. It is noted that not all variables reflecting the apparel industry
were included. Those missing were:
0.00
0.20
0.40
0.60
0.80
1.00
-2%
0%
2%
4%
6%
8%
10%
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Year
GDP Growth Rate Inflation
Interest Rate Unemployment Rate
Aggregate General Economic Indicator
0.00
0.20
0.40
0.60
0.80
1.00
40%
50%
60%
70%
80%
90%
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Year
Debt as a Percentage of GDP Aggregate General Economic Indicator
127

apparel exports because the volume is much less than and relatively insignificant to
imports;
industrial production index: clothing because it is continuously increasing and thus
offers little information;
apparel unit labour cost because its relationship with the economy is unclear. It is
often thought to be a predictor of inflation because it represents two-thirds of the total
costs to private US businesses. At the same time, however, rising unit labour costs
are associated with economic slowdown (Bureau of Labour Statistics, 2005); and,
CPI: apparel because this industry experiences deflation, likely due to goods
produced becoming increasingly cheaper offshore.

Figure 11.3: Apparel Industry Performance by Year
Table 11.2: Correlations to Aggregate Apparel Industry Indicator
Aggregate Apparel Industry Indicator
Personal Consumption Expenditures: Clothing 0.84
GDP: Clothing 0.85
Apparel Labour Productivity 0.49
Apparel Imports 0.83

11.3 Housing Market Indicator
To create an aggregate housing market indicator, one input (median months for a sale) and
one output (housing units started) were applied to DEA. Figure 11.4 and Figure 11.5 plot the
values of these factors which are graphed separately because their units and ranges are
0.00
0.20
0.40
0.60
0.80
1.00
-15%
-10%
-5%
0%
5%
10%
15%
20%
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Year
Personal Consumption Expenditures: Clothing GDP: Clothing
Apparel Labour Productivity Apparel Imports
Aggregate Apparel Industry Indicator
128

different. The strong correlations between each factor and the aggregate housing market
indicator (Table 11.3) proved that the reduction was acceptable.

Figure 11.4: Housing Market Performance and Months for a Sale by Year

Figure 11.5: Housing Market Performance and Construction by Year
Table 11.3: Correlations to Aggregate Housing Performance Indicator
Aggregate Housing Performance Indicator
Housing Units Started 0.97
Median Months for a Sale -0.83

11.4 Prices Indicator
To create an aggregate prices indicator, oil and cotton prices were treated as inputs and
combined with a unit output (Figure 11.6). Table 11.4 shows that positive correlations were
strong for oil price but weak for cotton price.
0.00
0.20
0.40
0.60
0.80
1.00
0
4
8
12
16
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Months
Year
Median Months for a Sale Aggregate Housing Market Indicator
0.00
0.20
0.40
0.60
0.80
1.00
0
500
1,000
1,500
2,000
2,500
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Units
Year
Housing Units Started Aggregate Housing Market Indicator
129

Figure 11.6: Prices by Year
Table 11.4: Correlations to Aggregate Prices Indicator
Aggregate Prices Indicator
Oil Price ($/bbl) -0.82
Cotton Price (cent/lb) -0.23

11.5 General Market Performance Indicator
Unlike the other categories, a general market indicator was created by taking the weighted
average of normalized NASDAQ and NYSE composite indices. The weights (0.45 and 0.55,
respectively) were based on the proportion of companies in this thesis traded on these
exchanges (Section 6.7).

Figure 11.7: General Market Performance by Year

11.6 All-Encompassing Market and Economic (ME) Indicator
In summary, 20 market and economic (ME) factors were simplified to 5 aggregate indicators
reflecting the general economy, apparel industry, commodity prices, housing market, and
market performance. The average of these 5 aggregates was then calculated for an all-
encompassing ME indicator (Figure 11.8).
0.00
0.20
0.40
0.60
0.80
1.00
0
20
40
60
80
100
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
DEA
Score
Year
Oil Price ($/bbl) Cotton Price (cent/lb) Aggregate Prices Indicator
0.00
0.20
0.40
0.60
0.80
1.00
0.00
0.20
0.40
0.60
0.80
1.00
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
Weighted
Average
Score
Normalized
Composite
Index
Year
Normalized NYSE Normalized NASDAQ Aggregate (Weighted) Market Indicator
130

Figure 11.8: All-Encompassing Market and Economic Indicator by Year
To test if this all-encompassing ME indicator was representative of overall market and
economic performance, its trend was matched to economic downturns during the time of
study. Between 1998 and 2008, there were three recessions:
1990-1991: triggered by the saving and loans crisis;
2001-2002: triggered by the dot-com bubble collapse and September 11 attacks; and,
2007-2009: triggered by the subprime mortgage crisis.
Because the minimums of the all-encompassing ME indicator curve coincided with
recessions, it was concluded that 20 factors were successfully reduced to one overall measure
of the market and economy consistent with reality.

11.7 Combining Market and Economic Factors with Financial and Managerial
Decision-Making Data
The hypothesis of this thesis states that the inclusion of managerial decision-making
outcomes as well as market and economic (ME) factors will enhance prediction that is based
solely on financial data. This inherently assumes that financial and managerial decision-
making data exclude external risks. If such is the case, there are three possibilities in which
ME factors can be integrated: 1. include them as non-discretionary inputs or outputs in the
first stage; 2. consider them with DEA scores or metrics before they are combined in a
second stage analysis; or 3. use them to calibrate the overall health score at the end.
Before exploring these non-trivial options, it must first be proven that the overall health of
each company differs from the all-encompassing ME indicator. This requires comparing
overall health by year to the all-encompassing ME indicator: if the curves are similar, then
0.00
0.20
0.40
0.60
0.80
1.00
1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008
Average
Score
Year
131

financial and managerial decision-making data do include ME effects, and thus strategies for
integrating them are unnecessary. This comparison is conducted in Section 12.7.

Figure 11.9: Possibilities of Adding Market and Economic Factors into DEA Models

Overall
Health
Second
Stage
SBM
Model
Metric IS
Income Statement
SBM Model
Incorporation of Market
and Economic Factors?
or
1. Include as inputs
or outputs for each
DMU in first stage?
2. Consider with
DEA scores before
second stage?
Balance Sheet - Assets
SBM Model
Metric BSA
Balance Sheet -
Liabilities
SBM Model
Metric BSL
Making
SBM Model
Metric MDM
or
3. Calibrate
overall health
score at the end?
132

12 Second Stage DEA Models: Results and Discussion
12.1 Motivation
Before delving into a second stage of analysis described in Section 8.2, the motivation was
reinforced. The alternative to creating metrics and combining them in a second stage was to
consider all financial and managerial decision-making variables at once; that is, a large set of
inputs and outputs was applied to a mega model. Each DMU (i.e. company and year) had
a value for each financial and managerial decision-making variable. For market and
economic (ME) factors, it was assigned values for each of these non-discretionary variables
as they match to its year.
In total, 35 inputs and 20 outputs were applied all at once to a DEA model (Table E.9).
Because the presence of non-discretionary variables disallowed the use of SBM for analysis,
a non-discretionary input-oriented BCC model was applied. Ninety-nine percent of DMUs
were identified as best performers. Even with the substitution of the all-encompassing ME
indicator for all the individual ME factors, 98% of DMUs were identified as healthiest. This
may have been due to too little variation in the data and/or building a model with a large
number of variables (relative to the number of DMUs) which increases dimensionality of the
mathematical program, and thereby landing more DMUs on the frontier. To illustrate,
consider the 7 input-oriented BCC models (with a dummy unit output) in Table 12.1.
The first model has one input, the second has two inputs and the third has three inputs,
and so on. Keeping the total number of DMUs constant (at 26), the number of DMUs on the
frontier increases with the number of inputs. Also, when the input value range is small (i.e.
lack of variation in data) between 1 and 3 the majority of DMUs appear on the frontier.
When the input value range is larger, e.g. 1 to 100 (Table 12.2), the number of DMUs on the
frontier decreases. And as before, Table 12.2 shows that the number of variables is
proportional to the number of DMUs on the frontier.

133

Table 12.1: Input-Oriented Models: Specifications and Results with Input Value Range between 1 and 3

Table 12.2: Input-Oriented Models: Specifications and Results with Input Value Range between 1 and
100

Input 1 Input 2 Input 3 Input 4 Input 5 Input 6 Input 7 Input 1 Inputs 1-2 Inputs 1-3 Inputs 1-4 Inputs 1-5 Inputs 1-6 Inputs 1-7
A 3 1 3 3 2 3 2 Yes Yes Yes Yes Yes Yes Yes
B 1 3 2 1 3 2 1 Yes Yes Yes Yes Yes Yes Yes
C 1 3 2 3 2 3 3 Yes Yes Yes Yes Yes Yes Yes
D 1 1 2 1 2 1 2 Yes Yes Yes Yes Yes Yes Yes
E 1 3 1 3 2 3 1 Yes Yes Yes Yes Yes Yes Yes
F 3 3 2 3 3 1 2 Yes Yes
G 3 3 3 3 3 3 2
H 2 3 1 1 3 2 1 Yes Yes Yes Yes Yes
I 2 3 1 2 2 2 1 Yes Yes Yes Yes Yes
J 2 1 2 2 3 3 3 Yes Yes Yes Yes Yes Yes Yes
K 1 3 3 3 3 3 1 Yes Yes Yes Yes Yes Yes Yes
L 1 1 2 2 1 1 3 Yes Yes Yes Yes Yes Yes Yes
M 1 3 2 2 1 3 2 Yes Yes Yes Yes Yes Yes Yes
N 3 3 1 1 1 1 2 Yes Yes Yes Yes Yes
O 1 2 2 3 2 1 1 Yes Yes Yes Yes Yes Yes Yes
P 3 2 2 3 3 3 3
Q 2 3 2 3 3 1 2 Yes Yes
R 1 3 2 2 1 3 1 Yes Yes Yes Yes Yes Yes Yes
S 3 2 2 1 1 3 3 Yes Yes Yes Yes
T 2 1 2 2 1 1 3 Yes Yes Yes Yes Yes Yes Yes
U 2 3 1 1 1 1 1 Yes Yes Yes Yes Yes
V 1 2 2 1 1 3 1 Yes Yes Yes Yes Yes Yes Yes
W 2 3 3 1 3 2 3 Yes Yes Yes Yes
X 1 2 2 3 2 3 1 Yes Yes Yes Yes Yes Yes Yes
Y 3 2 3 3 2 3 3
Z 3 3 1 2 1 3 2 Yes Yes Yes Yes Yes
14 14 19 21 21 23 23
DMU
Input Values DMUs on Frontier by Applying Model with:
Total Number of DMUs on Frontier
Input 1 Input 2 Input 3 Input 4 Input 5 Input 6 Input 7 Input 1 Inputs 1-2 Inputs 1-3 Inputs 1-4 Inputs 1-5 Inputs 1-6 Inputs 1-7
A 97 91 38 37 13 11 9 Yes Yes
B 26 67 87 48 5 31 58
C 29 51 6 89 36 80 40 Yes Yes Yes Yes Yes Yes Yes
D 51 67 52 57 13 67 66
E 25 45 73 85 63 40 51
F 96 65 80 33 50 81 69
G 60 73 12 17 49 95 60 Yes Yes Yes Yes
H 69 12 31 21 77 34 76 Yes Yes Yes Yes
I 6 42 20 1 3 15 13 Yes Yes Yes Yes Yes Yes Yes
J 23 54 80 51 33 34 42
K 80 28 36 98 22 41 52
L 78 67 22 60 43 75 57
M 9 32 33 22 85 92 57 Yes Yes Yes Yes Yes Yes
N 16 95 5 53 9 11 8 Yes Yes Yes Yes Yes
O 36 63 5 66 70 37 52 Yes Yes Yes Yes Yes
P 56 40 87 52 4 75 67 Yes Yes Yes
Q 100 79 71 73 17 75 57
R 71 77 41 82 87 67 46
S 26 3 45 78 39 29 39 Yes Yes Yes Yes Yes Yes
T 11 60 37 51 28 16 45 Yes Yes Yes Yes Yes
U 62 46 80 12 74 60 54 Yes Yes
V 17 72 83 49 79 77 58
W 31 51 70 90 40 36 21
X 43 86 66 93 69 97 5
Y 71 67 18 9 75 20 22
Z 81 2 2 60 65 72 77 Yes Yes Yes Yes Yes Yes
3 6 8 10 11 12 12
DMUs on Frontier by Applying Model with: Input Values
DMU
Total Number of DMUs on Frontier
134

12.2 Layered Scores and Second Stage Model Specifications
Because the layering technique provides more discrimination among DMUs and does not
require subjective zones, all second stage DEA models used scores derived from layering
results as variables. Specifically, the layer L corresponding to when a DMU was efficient
(on the frontier) after L peels (or runs) was translated into a novel layered score. For a
given metric with N layers, the layered score for a DMU was calculated as:
0 < Layered Score =
N + 1 - L
N
1
A DMU with the highest layered score of 1 was the healthiest within the sample; the lower
the layered score, the less healthy the DMU. In the instance where the DMU never appeared
on a frontier, it was given a layered score of 0. For example, BSL has 17 layers. Suppose a
DMU is on the frontier after the first run (L = 1). Its layered score would be
17 + 1 - 1
17
= 1 (the
healthiest). Similarly, if a DMU landed on the frontier after 6 peels (L = 6), its score would
be
17 + 1 - 6
17
= 0.706. The translation of layered results into layered scores was conducted for
each metric, and then correlated (Table 12.3). Consistent with correlations after the first run
(Table 10.1), this further supported the need to combine metrics in a second stage analysis.
Table 12.3: Correlation of Layered Scores

IS BSA BSL MDM
Income Statement (IS) 1 0.23 0.14 0.06
Balance Sheet Assets (BSA)

1 0.74 0.12
Balance Sheet Liabilities (BSL)

1 0.19
Managerial Decision-Making (MDM)

1

Many different combinations of metrics were applied to a second stage model. Some
had:
Financial data only (IS, BSA, BSL and/or CFO);
Financial and managerial decision-making data (IS, BSA, BSL and/or CFO, and
MDM); or
Financial and managerial decision-making data, and market and economic factors (IS,
BSA, BSL, CFO, MDM and ME).
Also, CFO (cash flow from operating activities) was sometimes added as a proxy because
negative values are a red flag of poor health.
135

All metrics in each second stage model were treated as outputs combined with a unit
input. All models were also non-oriented SBM except when the ME indicator was included,
in which case an output-oriented non-discretionary model was used. Finally, the scores
generated by second stage models reflected overall health of a DMU.

12.3 Classification by Zones
Second stage results based on classification by zones
27
are presented in Table 12.4.
Validating one part of the hypothesis, prediction using models with financial and managerial
decision-making improved over those with financial metrics only. It is noted that CFO had a
strong effect on prediction: its presence right-skewed the distribution (i.e. more DMUs with
lower scores), leading to zero Type I error but higher Type II error (Figure 12.1 versus Figure
12.2, and Figure 12.3 versus Figure 12.4). Furthermore, a model including the all-
encompassing ME indicator was also tested but produced the highest Type I error.
Table 12.4: Second Stage Model Prediction with Classification by Zones
%
IS, BSA, BSL IS, BSA, BSL, MDM IS, BSA, BSL,
CFO, MDM, ME Without CFO With CFO Without CFO With CFO
TP Rate 84.0 5.5 80.7 7.8 100.0
FP Rate 32.3 0 20.7 0 97.7
Type I Error 32.3 0 20.7 0 97.7
Type II Error 16.0 94.5 19.3 92.2 0
Success Rate 82.9 13.0 80.6 15.3 93.6
Unclassified 33.1 10.0 31.5 14.4 4.0

27
0.7.
136

Figure 12.1: Second Stage Results with IS, BSA and BSL

Figure 12.2: Second Stage Results with IS, BSA, BSL and
CFO

Figure 12.3: Second Stage Results with IS, BSA BSL and
MDM

Figure 12.4: Second Stage Results with IS, BSA, BSL,
CFO and MDM

12.4 Classification by Layering
Based on the second stage analysis of classification by zones (Section 12.3) which showed
that prediction was most effective with financial and managerial decision-making data but
0
25
50
75
100
125
150
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0
50
100
150
200
250
300
350
400
450
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0
25
50
75
100
125
150
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
0
50
100
150
200
250
300
350
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
137

not with market and economic factors, two combinations of metrics were further examined
by classification by layering:
1. IS, BSA, BSL and MDM
2. IS, BSA, BSL, and MDM with CFO

12.4.1 Second Stage Model with IS, BSA, BSL and MDM
Figure 12.5 graphs the count of actual active and bankrupt DMUs identified on the frontier in
each layer when predicting from one year back, on the primary y-axis (see Appendix F.6 for
two and three years back). The secondary y-axis plots the cumulative ratio of active and
bankrupt DMUs identified on the frontier to the total active and bankrupt DMUs in the
sample. Ideally, the Cumulative of Total Active DMUs curve (with green triangular
markers) would quickly increase while the Cumulative of Total Bankrupt DMUs curve
(with red square markers) would remain flat for as many layers as possible. Figure 12.6 is
the same as Figure 12.5, except that the primary y-axis is the ratio of active to bankrupt
DMUs on the frontier. As expected, as more layers were peeled, this ratio decreased.

Figure 12.5: Classification by Layer in Absolute Numbers from One Year Back

0%
20%
40%
60%
80%
100%
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
Cumulative % of DMUs
on Frontier
# DMUs on
Frontier
Layer
Active DMUs Bankrupt DMUs Cumulative of Total Active DMUs Cumulative of Total Bankrupt DMUs
138

Figure 12.6: Classification by Layer in Percentages from One Year Back
Figure 12.7 illustrate the common trade-offs between Type I and II errors, and active
and bankruptcy classifications, with each layer when predicting from one year back. Recall
that the goal was to create a model with a layer where both of its errors (red curves) are less
than (below) Altmans (blue lines). In fact, the point at which the error curves intersect
provides a measure of the models best performance, assuming the costs of Type I and II
errors are equal. Thus, a good prediction model would not have a point of intersection that
exceeds Altmans errors. Realistically, however, it would be successful provided that its
errors lie between Altmans benchmarks since Type I error is more costly than Type II error.

Figure 12.7: Prediction by Second Stage Model from One Year Back
A DMU was predicted bankrupt up to 1 year prior to filing Chapter 11.

Table 12.5 compares the performance, based on the intersections of error curves, between the
second stage model and individual metrics. The second stage model correctly classified 80%
of DMUs an improvement from Altmans Z-Score (58% accuracy from one year back).
0%
20%
40%
60%
80%
100%
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
on Frontier
% of DMU on
Frontier
Layer
% Active of Classified This Layer % Bankrupt of Classified This Layer
Cumulative of Total Active DMUs Cumulative of Total Bankrupt DMUs
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Layer
Type I Error
Type II Error
139

Generally, it was more accurate than each individual metric; any discrepancy was likely due
to the fact that individual metrics were computed from original data from annual reports
whereas second stage results were calculated from the layered scores of these metrics.
Table 12.5: Accuracy of Classification by Layering of Second Stage Model and Individual Metrics
One Year Back Two Years Back Three Years Back
Accuracy Error Layer Accuracy Error Layer Accuracy Error Layer
Second Stage 80% 20% 16 73% 27% 14 69% 31% 13, 14
IS 58% 42% 3 55% 45% 3 55% 45% 3
BSA 78% 22% 4, 5 72% 28% 4 70% 30% 3, 4
BSL 78% 22% 4 72% 28% 7, 8 70% 30% 7
MDM 65% 35% 2 60% 40% 2 61% 39% 2

To further compare the second stage model with the individual metrics, their Receiver
Operating Characteristic
28
(ROC) curves were plotted (Figure 12.8). All metrics and the
second stage model outperformed random guessing; however, the top two among them (i.e.
the second stage and BSL) appeared to leap-frog each other. To determine which was
better (since accuracies differed by only 2%), the area under each curve was calculated.
Because the area under the second stage model was identical to that of BSL (0.88), ROC
could not be used to choose the stronger performer. Thus, the evaluation of the models was
based on the accuracy of classification by layering, making the second stage model the best
predictor.

Figure 12.8: ROC Curves for One Year Back

28
An ROC curve is a graphical plot of sensitivity (or true positive rate) versus false positive rate (or 1 specificity or 1
true negative rate) for a binary classification system as its discrimination threshold is varied. It depicts the relative trade-offs
between true positive (benefits) and false positive (costs). Each prediction result represents one point in the ROC space.
The perfect prediction method would yield a point in the upper left corner (0, 1), representing 100% sensitivity (no false
negatives) and 100% specificity (no false positives). A completely random guess would give a point along the no-
discrimination diagonal (orange) line. Points above and below this diagonal represent good and poor classification,
respectively.
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
TP Rate
(Sensitivity)
FP Rate (1 - Specificity)
IS BSA BSL MDM Second Stage Random
140

12.4.2 Second Stage Model with IS, BSA, BSL and MDM with CFO
The inclusion of CFO as a metric to the previous model with IS, BSA, BSL and MDM
yielded 13 layers of analysis and did not perform as well as the second stage without CFO
(Table 12.6 and Figure 12.9).
Table 12.6: Second Stage Model with and without CFO Metric
One Year Back Two Years Back Three Years Back
Accuracy Error Layer Accuracy Error Layer Accuracy Error Layer
Without CFO 80% 20% 16 73% 27% 14 69% 31% 13, 14
With CFO 77% 23% 17 69% 31% 7 67% 33% 15

Figure 12.9: ROC Curves for One Year Back with CFO

12.5 Type I and Type II Error Trade-Off
As the ROC curves prove, all individual metrics and the second stage model were better than
random guessing but were not perfect. Because ROC curves do not reveal any information
on true negative or false negative (Type II error) instances, a comparative analysis of models
based on Type II errors was conducted. Figure 12.10 shows the trade-off between Type I and
II errors. If the objective is to minimize both simultaneously, then the second stage model
performed best since the intersection of its curve (coloured black) and the orange diagonal is
the lowest (0.2, 0.2) among the other metrics.
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
TP Rate
(Sensitivity)
FP Rate (1 - Specificity)
IS BSA BSL MDM Second Stage Random
141

Figure 12.10: Type I and II Errors Trade-Off
Further analysis of the Type I and II errors of the second stage model revealed that
the 20% error rate was a conservative estimate. In fact, for prediction one year back, the 6
bankrupt DMUs that were misclassified as active (Type I error) did not appear on the frontier
until the 8
th
peel and were clearly less healthy than DMUs that were classified as active
within the first few layers. Also, over a third of the active DMUs that were misclassified as
bankrupt (Type II) were companies that eventually filed for Chapter 11 (i.e. these DMUs
were out more than 3 years prior to bankruptcy).
Because attempts at improving a correct classification of 80% with data-mining
techniques (such as clustering) proved unsuccessful, literature was reviewed to see if other
studies have overcome the error trade-off problem. In short, varying the cut-off value can
increase specificity or sensitivity (Table 12.7): a more specific model (with a higher cut-off
value) has an increased risk of accepting false negatives (Type II error) while a more
sensitive model (with a lower cut-off value) has an increased risk of accepting false positives
(Type I error). In other words, there is no cut-off value with high sensitivity and high
specificity, but rather a continuum of risk at all values. Hence, the only options for
improving on errors are to introduce new variables that have not been considered in the
analysis, or to change the sample size (explored in Section 12.8).

0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
Type II
Error
Type I Error
IS
BSA
BSL
MDM
Second Stage
No Trade-Off
142

Table 12.7: Specificity and Sensitivity Trade-Off
Increasing Cut-Off Value Decreasing Cut-Off Value
Specificity
Sensitivity
True Positive
(Correctly predicting non-bankruptcy)

True Negative
(Correctly predicting bankruptcy)

False positive
(Incorrectly predicting non-bankruptcy)

False negative
(Incorrectly predicting bankruptcy)

12.6 Comparing the Layering and Non-Layering Approaches
Figure 12.11 presents the relationship between layering and non-layering (i.e. zoning)
techniques for the second stage model composed of metrics IS, BSA, BSL and MDM by
actual company state. The second stage layered scores
29
(y-axis) were calculated from results
after 25 peels whereas the non-layered scores (x-axis) were simply those generated by the
first DEA run. Each (x, y) pair is associated with one DMU. The dotted diagonal represents
a perfect correlation (y = x). The correlation between techniques was 0.82 and stronger when
the scores were higher; that is, there was less scatter when classification tended towards
active and greater variation in the bankrupt zone.
Figure 12.11 is also partitioned into 6 areas. The non-layered x-axis is separated into
3 zones: as a DMU with a non-layered score x was predicted bankrupt if 0 x 0.4 or active
if 0.7 x 1, or remained unclassified if 0.4 x < 0.7. Along the layered y-axis, there is a
single cut-off value that is determined by the optimal number of peels (i.e. when the second
stage model performed best), which in this case, was 16 for prediction one year back (Figure
12.7 and Table 12.6). Because the 16
th
layer corresponds to a layered score of 0.4, a DMU
with a layered score y was predicted bankrupt if y 0.4 or active if y > 0.4. It is noted that a
layered score is inversely related to the number of runs required for a given DMU to arrive at
the frontier. With these partitions, it was concluded that there was more variation in the
number of peels it took for DMUs to appear on the frontier when their non-layered scores
were low. Conversely, the higher the original non-layered score, the more reliable the

29
The second stage layered score for a DMU was calculated as y =
25 + 1 - L
25
where L is the layer corresponding to when a
DMU arrived on the frontier after L peels or runs.

143

measure and classification of Active was; in other words, one can be confident in a healthy
firm.

Figure 12.11: Correlation between Layering and Non-Layering Techniques
A DMU was predicted bankrupt up to 1 year prior to filing Chapter 11.
The strong correlation between non-layered and layered scores was also apparent in
prediction (Table 12.8). The evaluation results were nearly identical, but because of the
inability of the non-layering method to classify all DMUs, the layering technique was more
effective.
Table 12.8: Performance of Non-Layering and Layering Techniques
Evaluation
(%)
Layering Non-Layering:
Two Zones
Non-Layering:
Cut-Off of 0.8
Non-Layering:
Cut-Off of 0.7
Non-Layering:
Cut-Off of 0.6
TP Rate 80.3 80.7 37.7 54.6 66.9
FP Rate 20.0 20.7 0 4.0 20.0
Type I Error 20.0 20.7 0 4.0 20.0
Type II Error 19.7 19.3 62.3 45.4 33.1
Success Rate 80.3 80.6 39.9 56.1 67.3
Unclassified 0 31.5 0 0 0

Another interesting result was the existence of a lower boundary on the relationship
between layered (y) and non-layered scores (x) (Figure 12.11) which can be expressed as y
x
2
; this may imply that layered scores were more conservative estimates of health.

12.7 Comparison to All-Encompassing Market and Economic Indicator
Because the inclusion of the all-encompassing market and economic (ME) indicator
produced poor prediction results (Table 12.4), ME factors were excluded from second stage
0.00
0.20
0.40
0.60
0.80
1.00
0.00 0.20 0.40 0.60 0.80 1.00
Layered
Score
After
25 Runs
(y)
Non-layered Score (Score After First Run) (x)
Active Bankrupt
y = x
2
y = x
Classified by Zones
as Active
Classified by Zones
as Bankrupt
Unclassified by
Zones
Classified by
Optimal Layer
(16
th
) as Active
Classified by
Optimal Layer
(16
th
) as Bankrupt
144

models with the hypothesis that they are inherent in financial and managerial decision-
making data. To prove this, the second stage layered scores were averaged by year and
compared to the all-encompassing ME indicator (Figure 12.12).

Figure 12.12: Average Scores by Year
Assuming that the null hypothesis is that there is no significant difference between the two,
the similarity of their relationship was examined with T- and Mann Whitney U- tests,
Pearson and Spearman correlations, ANOVA, area under the curves, and Coefficients of
Divergence. These evaluations (Table 12.9) validated the hypothesis that there was no
significant difference between the means of the average second stage scores and the ME
indicator.
Table 12.9: Tests of Similarity between Second Stage Scores and ME Indicator

Evaluation Conclusion
Significance in Difference of Means
Mann-Whitney U Test 0.29
No significant difference
between means of scores
T-Test 0.77
ANOVA 0.77
Correlation
Pearson 0.62 Weak correlation
between scores Spearman 0.50
Difference in Area
Area under the Curves 3.6% No significant difference
Coefficient of Divergence (Wongphatarakul et al., 1998)
Coefficient of Divergence 0.15 Similarity

While their correlation was weak, it was the matching of the direction of change from year to
year that was more important than the magnitude of change itself. In fact, the average second
stage score and ME indicator were in agreement 10 out of 12 years (Table 12.10). Thus, it
was concluded that financial and managerial data have ME factors inherent in their values.
0.0
0.2
0.4
0.6
0.8
1.0
0.30
0.40
0.50
0.60
0.70
1997 1999 2001 2003 2005 2007 2009
Market and
Economic
Indicator Score
Second Stage
Average
DEA Score
Year
Second Stage Market and Economic Indicator
145

Table 12.10: Direction of Change from Year to Year
Year Second Stage Market and Economic Indicator
1997-1998 Down Down
1998-1999 Up Up
1999-2000 Up Down
2000-2001 Down Down
2001-2002 Down Down
2002-2003 Down Up
2003-2004 Down Down
2004-2005 Up Up
2005-2006 Down Down
2006-2007 Down Down
2007-2008 Down Down
2008-2009 Up Up

12.8 Improving Prediction by Changing Sample Size
As discussed in Section 12.5, the only options for improving Type I and II errors are to
introduce new variables or change the sample size. Because trade data and CFO were
collected but unused in the formation of second stage results, they were later combined with
second stage scores and applied as variables to other data mining techniques in an attempt to
improve prediction. However, these efforts were futile.
Until this point, with all companies studied between 1996 to 2009, the ratio of active
to bankrupt companies was 61 to 18 (3.4 to 1), and the ratio of active to bankrupt DMUs
(company and year) was 651 to 50 (13 to 1). What happens when this ratio is reduced?
Figure 12.13, Figure 12.14, Figure 12.15 and Figure 12.16 plot the sensitivity, specificity,
and Type I and Type II errors of the second stage model as a function of cut-off value, while
decreasing the ratio of active to bankrupt firms from 3.4:1 to 3:1, 2:1 and 1:1, respectively.
Here, each DMU represented a company only with a score averaged of all years in which
data were available for the company. The companies selected for each subset were not
random but the best and worst firms as determined by their average layered scores. This
practice was justified as in theory, it is important to build a database of the best and worst
companies such that new data (a new company) can be judged against these benchmarks.
As the ratio of active to bankrupt firms dropped, the sensitivity and specificity
increased while the Type I and II errors decreased. In fact, when the ratio was either 2:1 or
1:1 with a cut-off value of 0.6, and between 0.60 and 0.70, respectively, there was perfect
classification. This result was consistent with the common practice taken by Altman and
other studies that model with an equal number of active and bankrupt companies (known to
146

improve prediction results) despite not being representative of reality where the ratio of
active to bankrupt firms is much higher. In other words, the second stage model developed
in this thesis was a less biased approach given that it maintained a more natural ratio of
active to bankrupt companies.

Figure 12.13: All Companies (3.4-to-1 Ratio)

Figure 12.14: 3-to-1 Ratio



12.9 Translating DEA Results into Targets of Improvement
A major advantage of DEA technology is that it provides tangible targets for improvement.
Recall that an inefficient DMU can be improved by referring its inefficient behaviour to
the efficient frontier formed by E
o
, the reference set of DMU
o
composed of efficient DMUs.
This improvement is an SBM projection to the point
on the frontier where:

) ..., , 1 (
) ..., , 1 (
*
s r s y y
m i s x x
r ro ro
-*
i io io
= + =
= =
+
(12.1)
0%
20%
40%
60%
80%
100%
0.00 0.20 0.40 0.60 0.80 1.00
Cut-off Value
Sensitivity (TP Rate) Specificity
Type I Error (FP Rate) Type II Error
0%
20%
40%
60%
80%
100%
0.00 0.20 0.40 0.60 0.80 1.00
Cut-off Value
0%
20%
40%
60%
80%
100%
0.00 0.20 0.40 0.60 0.80 1.00
Cut-off Value
0%
20%
40%
60%
80%
100%
0.00 0.20 0.40 0.60 0.80 1.00
Cut-off Value
147

are the coordinates of a virtual linear composite DMU (i.e. DMU

i
i
where DMU
i
s
are efficient and
i
are proportionality weights for DMU
i
) used to evaluate the performance of
DMU
o
. It represents the target for efficient production that DMU
o
strives for (Cooper et al.,
2007). s
+*
are output shortfalls while s
-*
are inputs which must be reduced and augmented,
respectively, to improve efficiency.
As an example, Figure 12.17 presents the second stage projections of an active firm
(AEO) and bankrupt firm (FTUS) based on non-layered results (i.e. after one run). These
DMUs, which were not on the frontier because their second stage scores were less than 1,
show that the lower the score, the greater the possibilities for expansion of outputs for
improvement. In fact, the second stage projections are a guide for which aspects of the
company, as represented by individual metrics, should be targeted.

Figure 12.17: Projections for Metrics in the Second Stage for Selected DMUs
For instance, Figure 12.18 presents the projections for the income statement (IS) metric. As
seen, in the years prior to bankruptcy, the COGS and SGA of FTUS could be reduced by as
much as 20%. In addition, FTUSs net income was significantly less than the healthiest
companies. This exercise can be applied to each metric to learn which specific variables can
be improved upon. There is also room for improvement for active DMUs that are not among

AEO 1997 A 0.30 50% 60% 89% 1000%
AEO 1998 A 0.50 200% 60% 113% 20%
AEO 1999 A 0.54 200% 60% 55% 20%
AEO 2000 A 0.75 50% 33% 31% 20%
AEO 2001 A 0.71 100% 14% 31% 20%
AEO 2002 A 0.81 50% 0% 21% 20%
AEO 2003 A 0.76 20% 0% 6% 100%
AEO 2004 A 0.72 20% 14% 21% 100%
AEO 2005 A 0.86 0% 0% 13% 50%
AEO 2006 A 0.76 20% 0% 6% 100%
AEO 2007 A 0.89 0% 0% 0% 50%
AEO 2008 A 0.89 0% 0% 0% 50%
AEO 2009 A 0.89 0% 0% 0% 50%

FTUS 1996 A 0.19 100% 167% 750% 1000%
FTUS 1997 A 0.15 500% 60% 1000% 100%
FTUS 1998 A 0.17 100% 60% 1000% 200%
FTUS 1999 A 0.28 100% 60% 750% 100%
FTUS 2000 A 0.28 500% 167% 325% 20%
FTUS 2001 B-3 0.50 100% 100% 183% 20%
FTUS 2002 B-2 0.46 20% 167% 240% 50%
FTUS 2003 B-1 0.23 50% 300% 467% 500%
FTUS 2004 B 0.17 1000% 100% 467% 1000%

IS (Output)
Change
BSA (Output)
Change
BSL (Output)
Change
MDM (Output)
Change
Company Year State
Second Stage
Score
Active
Company
Bankrupt
Company
Greater possible
expansion of outputs
for improvement
148

the healthiest group. These projections not only provide valuable bankruptcy preventative
measures, but also profitability strategies.

Figure 12.18: Projections for IS Metric Variables for Selected DMUs
Figure 12.19 Figure 12.53 present the recommended percent changes for all
variables of each metric: for inputs, these projections were reductions while for outputs, they
were expansions. These histograms focus primarily on bankrupt DMUs. For the results of
active companies, please refer to Appendix F.7.
On average, the greatest opportunities for improvement on the income statement for
bankrupt DMUs were to reduce net interest expense and income tax expense, as well as to
increase net income (Figure 12.19 Figure 12.24).
Active
Company
Bankrupt
Company
Possible reduction of
inputs for improvement
Possible expansion
of outputs for
improvement
Projection Change Projection Change Projection Change

AEO 1997 A 0.29 175 -5% 127 0% 27 361%
AEO 1998 A 0.60 218 -1% 146 0% 35 77%
AEO 1999 A 0.84 293 0% 199 0% 73 35%
AEO 2000 A 0.91 398 0% 272 0% 107 18%
AEO 2001 A 0.86 657 0% 266 0% 122 31%
AEO 2002 A 0.86 825 0% 339 0% 141 34%
AEO 2003 A 0.83 842 0% 328 0% 124 41%
AEO 2004 A 0.67 886 0% 355 0% 119 99%
AEO 2005 A 1 866 0% 592 0% 213 0%
AEO 2006 A 0.98 1244 0% 539 0% 307 4%
AEO 2007 A 1 1454 0% 666 0% 387 0%
AEO 2008 A 1 1632 0% 715 0% 400 0%
AEO 2009 A 1 1815 0% 734 0% 179 0%

FTUS 1996 A 0.03 117 0% 56 0% 71 1000%
FTUS 1997 A 0.00 163 -9% 85 -5% 122 1000%
FTUS 1998 A 0.00 204 0% 87 0% 93 1000%
FTUS 1999 A 0.04 222 0% 99 0% 78 1000%
FTUS 2000 A 0.26 271 0% 120 0% 74 497%
FTUS 2001 B-3 0.44 358 0% 154 0% 72 238%
FTUS 2002 B-2 0.00 382 0% 187 0% 39 1000%
FTUS 2003 B-1 0.00 357 0% 170 -12% 52 1000%
FTUS 2004 B 0.00 276 -19% 158 -12% 115 1000%

NI (Output)
Company Year State
IS
Score
COGS (Input) SGA (Input)
149

Figure 12.19: IS Projections for COGS
Mean = 2.2% 4.9%; Median = 0%
Figure 12.20: IS Projections for SGA
Mean = 7% 11%; Median = 0%

Figure 12.21: IS Projections for Net Interest
Expense
Mean = 30% 35%; Median = 10%;
Figure 12.22: IS Projections for Income Tax
Expense
Mean = 25% 38%; Median = 0%

0
5
10
15
20
25
30
35
F
r
e
q
u
e
n
c
y
Possible Reduction for Improvement
Bankrupt
0
5
10
15
20
25
30
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
35
F
r
e
q
u
e
n
c
y
Bankrupt
150

Figure 12.23: IS Projections for Revenue
Mean = 0.1% 0.4%; Median = 0%

Figure 12.24: IS Projections for Net Income
Mean = 710% 390%; Median = 999%

Within asset categories on the balance sheet, the average bankrupt DMUs could have
reduced accounts receivable, inventories, goodwill, PPE and total assets, and increased cash,
retained earnings and shareholders equity (Figure 12.25 Figure 12.35).
Figure 12.25: BSA Projections for Marketable
Securities
Mean = 4% 20%; Median = 0%

Figure 12.26: BSA Projections for Accounts
Receivable
Mean = 90% 28%; Median = 100%

0
5
10
15
20
25
30
35
40
45
50
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
Possible Expansion for Improvement
(By Factor)
Bankrupt
0
5
10
15
20
25
30
35
40
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
10
20
30
40
F
r
e
q
u
e
n
c
y
Bankrupt
0
10
20
30
40
50
F
r
e
q
u
e
n
c
y
Bankrupt
151

Figure 12.27: BSA Projections for Inventories
Mean = 54% 24%; Median = 62%

Figure 12.28: BSA Projections for Current Assets
Mean = 2.5% 6.3%; Median = 0%

Figure 12.29: BSA Projections for Long-Term
Investment in Securities
Mean = 8% 25%; Median = 0%

Figure 12.30: BSA Projections for Net PPE
Mean = 21% 30%; Median = 0%

0
5
10
15
20
F
r
e
q
u
e
n
c
y
Bankrupt
0
10
20
30
40
F
r
e
q
u
e
n
c
y
Bankrupt
0
10
20
30
40
50
F
r
e
q
u
e
n
c
y
Bankrupt
0
10
20
30
F
r
e
q
u
e
n
c
y
Bankrupt
152

Figure 12.31: BSA Projections for Goodwill
Mean = 42% 48%; Median = 0%

Figure 12.32: BSA Projections for Total Assets
Mean = 16% 13%; Median = 18%

Figure 12.33: BSA Projections for Cash
Mean = 620% 420%; Median = 990%

Figure 12.34: BSA Projections for Retained
Earnings
Mean = 610% 430%; Median = 890%

0
10
20
30
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
35
40
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
20
25
30
35
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
153

Figure 12.35: BSA Projections for Shareholders Equity
Mean = 210% 360%; Median = 24%

With respect to liabilities on the balance sheet, the average bankrupt DMU could have
reduced notes payable and short-term debt, long-term debt, accounts payable, current
maturities, current liabilities, total liabilities, and increased retained earnings and
shareholders equity (Figure 12.36 Figure 12.43).
Figure 12.36: BSL Projections for Accounts
Payable
Mean = 36% 29%; Median = 37%

Figure 12.37: BSL Projections for Notes Payable
and Short-Term Debt
Mean = 44% 50%; Median = 0%

0
5
10
15
20
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
2
4
6
8
10
12
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
F
r
e
q
u
e
n
c
y
Bankrupt
154

Figure 12.38: BSL Projections for Current
Maturities of Long-Term Debt
Mean = 44% 33%; Median = 46%

Figure 12.39: BSL Projections for Current
Liabilities
Mean = 40% 30%; Median = 36%

Figure 12.40: BSL Projections for Long-Term
Debt
Mean = 67% 48%; Median = 100%

Figure 12.41: BSL Projections for Total Liabilities
Mean = 43% 32%; Median = 43%

0
2
4
6
8
10
12
F
r
e
q
u
e
n
c
y
Bankrupt
0
3
6
9
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
35
F
r
e
q
u
e
n
c
y
Bankrupt
0
2
4
6
8
10
F
r
e
q
u
e
n
c
y
Bankrupt
155

Figure 12.42: BSL Projections for Retained
Earnings
Mean = 710% 380%; Median = 1000%

Figure 12.43: BSL Projections for Shareholders
Equity
Mean = 380% 350%; Median = 250%
Opportunities for improvement with respect to managerial decision-making for
bankrupt DMUs were to reduce management turnover, legal proceedings, related party
transactions and retirement plans (Figure 12.44 Figure 12.49).
Figure 12.44: MDM Projections for Related Party
Transactions
Mean = 32% 47%; Median = 0%
Figure 12.45: MDM Projections for Auditors
Opinion
Mean = 0% 0%; Median = 0%

0
5
10
15
20
25
30
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
20
25
30
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
35
40
45
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
156

Figure 12.46: MDM Projections for Change of
Auditor
Mean = 10% 30%; Median = 0%

Figure 12.47: MDM Projections for Management
Turnover
Mean = 48% 49%; Median = 50%

Figure 12.48: MDM Projections for Legal
Proceedings
Mean = 45% 47%; Median = 0%

Figure 12.49: MDM Projections for Retirement
Plan
Mean = 27% 45%; Median = 0%

Finally, in terms of overall health (second stage) of bankrupt DMUS, all metrics could
have been improved upon (Figure 12.50 Figure 12.53).
0
5
10
15
20
25
30
35
40
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
0
5
10
15
20
25
30
0% 50% 100%
F
r
e
q
u
e
n
c
y
Bankrupt
157

Figure 12.50: Second Stage Projections for IS
Metric
Mean = 220% 280%; Median = 100%

Figure 12.51: Second Stage Projections for BSA
Metric
Mean = 360% 400%; Median = 100%

Figure 12.52: Second Stage Projections for BSL
Metric
Mean = 280% 330%; Median = 160%
Figure 12.53: Second Stage Projections for MDM
Metric
Mean = 300% 410%; Median = 50%

0
5
10
15
20
25
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
20
25
30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
20
25
30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
0
5
10
15
20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
Bankrupt
158

13 Prediction and Validation of the Second Stage DEA Model
13.1 Estimating the Probability of Bankruptcy
Like Altmans Z-Score and most studies, results from the layering technique provide
dichotomous predictions of bankruptcy (i.e. yes or no) but not probabilities. Knowing
the chance of failure is valuable because it associates a level of confidence or warning to
each DMU; that is, one can answer the question: what is the probability of a DMU going
bankrupt given a particular second stage layered score?
A histogram of the second stage layered scores shows that as the scores increased,
fewer bankrupt DMUs appeared on the frontier (Figure 13.1). The cumulative number of
active and bankrupt DMUs on the frontier, starting from the last layer (i.e. when the score is
zero), were counted, from which the actual probabilities of bankruptcy and non-bankruptcy
were then computed (Table 13.1). As expected, the higher the score (which is inversely
related to the layer that a DMU appears on the frontier), the less probability of bankruptcy.
With actual discrete probabilities for comparison, the scores were translated into continuous
probabilities with the cumulative distribution function (CDF) of the standard normal
distribution and a second order polynomial.

Figure 13.1: Distribution of Second Stage Layered Scores
Mean = 0.55 0.23; Median = 0.56; Kurtosis = -0.39; Skewness = 0.56

0
10
20
30
40
50
60
0
.
0
0
0
.
0
4
0
.
0
8
0
.
1
2
0
.
1
6
0
.
2
0
0
.
2
4
0
.
2
8
0
.
3
2
0
.
3
6
0
.
4
0
0
.
4
4
0
.
4
8
0
.
5
2
0
.
5
6
0
.
6
0
0
.
6
4
0
.
6
8
0
.
7
2
0
.
7
6
0
.
8
0
0
.
8
4
0
.
8
8
0
.
9
2
0
.
9
6
1
.
0
0
F
r
e
q
u
e
n
c
y
Layered Score
Bankrupt
Active
159

Table 13.1: Probabilities of Bankruptcy (B) and Non-Bankruptcy (NB)
A DMU is classified bankrupt up to 3 years prior to filing Chapter 11.
Layer Score
Actual Estimates
Count of
DMUs on
Frontier
Cumulative Count of
DMUs on Frontier P(NB) P(B)
Cumulative
Distribution
Function
Second order
Polynomial
Function
Total NB B Total NB B P(NB) P(B) P(NB) P(B)
26 0.00 - - - - - - - - 5.1% 94.9% 60.8% 39.2%
25 0.04 6 4 2 6 4 2 66.7% 33.3% 6.6% 93.4% 63.9% 36.1%
24 0.08 11 7 4 17 11 6 64.7% 35.3% 8.5% 91.5% 66.8% 33.2%
23 0.12 10 6 4 27 17 10 63.0% 37.0% 10.7% 89.3% 69.5% 30.5%
22 0.16 10 9 1 37 26 11 70.3% 29.7% 13.3% 86.7% 72.1% 27.9%
21 0.20 16 14 2 53 40 13 75.5% 24.5% 16.3% 83.7% 74.6% 25.4%
20 0.24 18 18 0 71 58 13 81.7% 18.3% 19.8% 80.2% 76.9% 23.1%
19 0.28 18 15 3 89 73 16 82.0% 18.0% 23.6% 76.4% 79.0% 21.0%
18 0.32 23 19 4 112 92 20 82.1% 17.9% 27.8% 72.2% 81.0% 19.0%
17 0.36 32 29 3 144 121 23 84.0% 16.0% 32.4% 67.6% 82.9% 17.1%
16 0.40 30 28 2 174 149 25 85.6% 14.4% 37.2% 62.8% 84.6% 15.4%
15 0.44 29 27 2 203 176 27 86.7% 13.3% 42.2% 57.8% 86.1% 13.9%
14 0.48 45 42 3 248 218 30 87.9% 12.1% 47.4% 52.6% 87.5% 12.5%
13 0.52 48 43 5 296 261 35 88.2% 11.8% 52.6% 47.4% 88.8% 11.2%
12 0.56 52 48 4 348 309 39 88.8% 11.2% 57.8% 42.2% 89.9% 10.1%
11 0.60 36 35 1 384 344 40 89.6% 10.4% 62.8% 37.2% 90.8% 9.2%
10 0.64 49 49 0 433 393 40 90.8% 9.2% 67.6% 32.4% 91.7% 8.3%
9 0.68 46 44 2 479 437 42 91.2% 8.8% 72.2% 27.8% 92.3% 7.7%
8 0.72 56 54 2 535 491 44 91.8% 8.2% 76.4% 23.6% 92.8% 7.2%
7 0.76 39 39 0 574 530 44 92.3% 7.7% 80.2% 19.8% 93.2% 6.8%
6 0.80 38 38 0 612 568 44 92.8% 7.2% 83.7% 16.3% 93.4% 6.6%
5 0.84 24 24 0 636 592 44 93.1% 6.9% 86.7% 13.3% 93.5% 6.5%
4 0.88 23 23 0 659 615 44 93.3% 6.7% 89.3% 10.7% 93.4% 6.6%
3 0.92 19 19 0 678 634 44 93.5% 6.5% 91.5% 8.5% 93.1% 6.9%
2 0.96 9 9 0 687 643 44 93.6% 6.4% 93.4% 6.6% 92.7% 7.3%
1 1.00 1 1 0 688 644 44 93.6% 6.4% 94.9% 5.1% 92.2% 7.8%

Recall that the probability density function (PDF) of a random variable describes the
relative frequencies of different values for that random variable. The PDF for the normal
distribution is:
fx; ,
2
=
1
2
2
e
-(x - )
2
/(2
2
)

where and
2
are the mean and variance of the distribution, respectively. The CDF of the
standard normal distribution, which describes the probability of a random variable falling in
the interval (-, x] is:
x=
1
2
e
-t
2
/2
X
-
dt
160

Assuming that the second stage layered scores were normally distributed, each layered score
was first transformed to a standard normal score
30
, and then applied to the CDF. As
presented in Table 13.1, the estimated probabilities were consistent with the actual
probabilities in terms of direction, but had a greater range (5-95% and 6-33%, respectively).
A second order polynomial function was also used to predict the probability of
bankruptcy by fitting the second set of layered scores to:
y = 0.47x
2
- 0.78x + 0.39
where y is the probability of bankruptcy and x is second stage layered score. This function
produced estimated probabilities that matched closer to actual probabilities. The information
in Table 13.1 is also represented graphically in Figure 13.2, which plots the relationship
between second stage layered score and the probability of bankruptcy, from three years back.

Figure 13.2: Probability of Bankruptcy

13.2 Reliability and Validity of the Second Stage Model
Because all previous results considered data from 1996 to 2009, in order to test the reliability
of the two-stage layered DEA model, the entire process of creating metrics (with the same
variables outlined in Sections 10.1, 10.2 and 10.4) and combining them in a second step was
repeated 5 additional times with smaller datasets spanning 1996 to year Y, where Y ranged
from 2004 to 2008 (Figure 13.3).

30
If X is normal with mean and standard deviation , then
has the standard normal distribution.

y = 0.47x
2
- 0.78x + 0.39
R = 0.95
0%
20%
40%
60%
80%
100%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Second Stage Layered Score
CDF Standard Normal Estimate
Actual Data
Second Order Polynomial
161

a: Use data from 1996 to 2009 to
predict health in 2010 and 2011
b: Use data from 1996 to 2008 to
c: Use data from 1996 to 2007 to
d: Use data from 1996 to 2006 to
e: Use data from 1996 to 2005 to
f: Use data from 1996 to 2004 to
Figure 13.3: Time Windows for Model Testing

The results from classification by layering from one year back are summarized in
Table 13.2, based on plots presented in Appendix F.8. The accuracy of the second stage
model slightly decreased as the time window shortened. This was expected as DEA
performs better with more DMUs. Nonetheless, because the model was able to predict all
DMUS with an accuracy between 75-80%, this suggests that the second stage model was
reliable, and thereby implies that the selection of variables for the metrics was robust.
Table 13.2: Accuracy of Classification by Layering of Second Stage Model from One Year Back
Time Window Total Number of DMUs Total Number of Peels Accuracy Error Layer
1996 2009 701 (651 Active, 50 Bankrupt) 26 80% 20% 16
1996 2008 584 (543 Active, 41 Bankrupt) 23 76% 24% 14
1996 2007 532 (492 Active, 40 Bankrupt) 22 76% 24% 13, 14
1996 2006 476 (439 Active, 37 Bankrupt) 21 76% 24% 13
1996 2005 409 (379 Active, 30 Bankrupt ) 19 75% 25% 11, 12
1996 2004 357 (331 Active, 26 Bankrupt) 18 77% 23% 11, 12

As in the previous section (13.1), the probability of bankruptcy was estimated by
fitting the second set of layered scores to a second order polynomial function. The functions
based on the datasets with different time windows are presented in Table 13.3. The graphs
comparing the actual probabilities of bankruptcy to the estimate are presented in Appendix
F.9.
Table 13.3: Second Order Polynomial Functions Estimating the Probability of Bankruptcy
A DMU was classified bankrupt up to a year prior to filing Chapter 11.
Time Window Equation Probability of Bankruptcy (y) from Second Stage Layered Score (x)
1996 2009 a y = 0.54x
2
- 0.81x + 0.32
1996 2008 b y = 0.63x
2
- 0.91x + 0.34
1996 2007 c y = 0.24x
2
- 0.41x + 0.20
1996 2006 d y = 0.16x
2
- 0.31x + 0.17
1996 2005 e y = 0.05x
2
- 0.12x + 0.10
1996 2004 f y = -0.24x
2
+ 0.24x

These datasets were then used to predict health over the next two years (i.e. years Y +
1 and Y + 2). The effectiveness of the prediction models are discussed in the following
1996 2004 2005 2006 2007 2008 2009 2010 2011
a
b
c
d
e
f
162

sections. In general, the reliability of the models decreased as the time frame was shortened
because of the reduction of DMUs resulting in lower accuracy.

13.2.1 Prediction of Health after 2009 based on Data from 1996 to 2009
Data for 39 companies (all active) were available in 2009, lending to 39 second stage layered
scores and corresponding probabilities for 2009 based on a second order polynomial function
(Equation a in Table 13.3). In this year, 23 of these firms had scores greater than 0.6 or
probability of bankruptcy at less than 5%. This low chance was consistent with the fact that
all these companies remained active through 2010 and 2011. In contrast, there were 3
companies with scores less than 0.3 or an 11% chance of bankruptcy. In 2010 and 2011, one
of these 3 companies declared bankruptcy; another entered a transitional period; and one
remained active. In 2009, 13 companies had scores between 0.3 and 0.6, or a chance of
bankruptcy between 5% and 11%. In 2010 and 2011, most of these companies reached new
credit agreements, closed stores, merged with other companies and/or went private showing
that they were in a less stable state with a higher risk of bankruptcy compared to those with
higher scores.

13.2.2 Prediction of Health after 2008 based on 1996-2008 Data
Data for 52 companies were available in 2008, lending to 52 second stage layered scores and
corresponding probabilities for 2008 based on a second order polynomial function (Equation
b in Table 13.3). In this year, 29 active firms had scores greater than 0.6 or probability of
bankruptcy at less than 5%. This low likelihood was consistent in that 27 of these companies
remained active through 2009 and 2010. However, 2 of these firms filed for Chapter 11
protection at the onset of recession. In addition, there were 8 companies with scores less than
0.4 or a 6% chance of bankruptcy. In 2009 and 2010, two of these companies declared
bankruptcy; one went private; one changed its name; one emerged from bankruptcy; and
three remained active. Also, in 2008, there were 15 companies with scores between 0.4 and
0.6, or a chance of bankruptcy between 5% and 6%. In 2009 and 2010, two-thirds of these
companies remained active, while the others were in less stable health as evident by new
credit agreements and store closings.

163

c in Table 13.3). In this year, 35 firms had scores greater than 0.6 or probability of
bankruptcy at less than 5%. This low chance was consistent with the fact that all but one of
these companies remained active through 2008 and 2009. On the other hand, 6 companies
had scores less than 0.3 or a 10% chance of bankruptcy. In 2008 and 2009, 3 of these
companies declared bankruptcy; one emerged from bankruptcy; one went private; and one
remained active. Furthermore, in 2007, there were 15 companies with scores between 0.3
and 0.6, or a chance of bankruptcy between 5% and 6%. In 2008 and 2009, over two-thirds
of these companies remained active, while two firms filed for Chapter 11 protection at the
onset of recession.

d in Table 13.3). In this year, 30 firms had scores greater than 0.6 or probability of
bankruptcy at less than 5%. all these companies remained active through 2007 and 2008.
In contrast, 6 companies had scores less than 0.3 or a 9% chance of bankruptcy. In 2007 and
2008, only two of these companies actually declared bankruptcy while the others remained
active. Moreover, in 2006, 14 companies had scores between 0.3 and 0.6, or a chance of
bankruptcy between 5% and 9%. In 2007 and 2008, two firms filed for bankruptcy while the
other remained active.

e in Table 13.3). In this year, 26 firms had scores greater than 0.6 or probability of
bankruptcy at 4% all of which remained active through 2006 and 2007. In contrast, 5
companies had scores less than 0.3 or a 6% chance of bankruptcy. In 2006 and 2007, only
one of these companies actually declared bankruptcy while the others remained active.
164

Furthermore, in 2006, 16 companies had scores between 0.3 and 0.6, or a chance of
bankruptcy between 4% and 6%. In 2006 and 2007, one firm filed for Chapter 11 protection
while the others remained active.

f in Table 13.3). Because the function does not increase monotonically (Figure F.61), the
estimation for the probability of bankruptcy based on data between 1996 and 2004
exclusively, could not be used with reliability.

165

14 Conclusions
14.1 Summary of Major Contributions
In this work, 85 companies in the American retail-apparel industry were studied. These
public companies have been traded for some continuous period of time between 1996 and
2009. Two types of information were collected for each company for each year an annual
report was available: financial (i.e. numerical variables from balance sheet, income
statement and cash flow statement) and managerial decision-making (i.e. categorical
variables from Notes, MD&A and Auditors Report). Data collection was followed by an
extensive survey of bankruptcy and fraud literature, and the profiling of the retail-apparel
industry. Next, the exploration of financial and managerial decision-making data involved
correlating them with one another as well as with key market and economic factors. Results
showed that financial, managerial decision-making, and market and economic data were
independent and individually were poor predictors of bankruptcy. It was therefore
hypothesized that the inclusion of variables that reflect managerial decision-making, and
market and economic factors, enhance the predictive power of current mathematical models
that consider financial data exclusively.
With a unique and comprehensive dataset, metrics based on different aspects of the
annual report were created then combined with a Data Envelopment Analysis (DEA) model
and modified layering technique. This two-stage approach proved to be an effective
classification tool, separating companies with a high risk of bankruptcy from those that were
healthy, with 80% accuracy when predicting from one year back. This 22% improvement
over Altmans industry-standard model was attributed to prediction with financial and
managerial decision-making data but not with market and economic factors which were
shown to be inherent in the values of the former data types.
To summarize, the development of a DEA-based model led to several contributions.
A rich database for each company was created with managerial decision-making
information collected from the Notes, MD&A and Auditors Report. The use of this
data in a mathematical model was novel and will act as a model for others to follow.
Novel metrics that reflect different aspects of the annual report were created using
normal SBM DEA models. This also offers a new direction for further research.
166

The two-stage model and modified layering technique provided a probability of
bankruptcy in addition to a dichotomous prediction. It also offered targets for
improvements which can be viewed as profitability strategies and/or preventative
measures to help firms out of a high risk of bankruptcy. By providing probabilities,
the human decision making process is enhanced when compared to a yes/no answer.
The methodology was designed to be easily adapted for analyses of other industries
(Figure 14.1). For example, if one were to study the manufacturing industry, the
same metrics could be created and then combined in a second stage model for an
overall measure of company health. The selection of variables for each metric would
be at the discretion of the researcher but the philosophy of this thesis was to consider
those that have an effect on company state, as determined by correlations and tests of
similarly. In the manufacturing industry, a more capital-intensive industry than retail-
apparel, it would be anticipated that PPE would have a strong influence on the asset
metric. There may also be more off-balance sheet liabilities to look for in the Notes
such as special purpose entities (SPEs). Moreover, the cash flow statement and trade
data may provide valuable information as metrics, if added. Different economic
factors that pertain to the manufacturing industry would have to be considered. This
industry would likely be more sensitive to the performance of the economy,
especially as more US firms invest in off-shore operations.
Finally, the contributions of this thesis have practical applications, with economic and
social benefits. For creditors, bankruptcy prediction improves risk assessment while for
owners, it would buy extra time to secure more financing or improve current operations to
avoid failure altogether. For investors, the model can also be used to identify active and
bankrupt companies, helping one to invest in healthy companies or to short unhealthy firms.
167

Figure 14.1: Summary of Methodology as Applicable to All Industries

14.2 Recommendations for Future Work
Several areas for future research and improvement in both bankruptcy prediction and DEA
arise from this work. Opportunities include:
Adding information that is excluded from the annual reports such as:
o The number of employees and/or layoffs (i.e. turnover) since human capital is
pertinent in the services industry;
o The number of new store openings and closings;
o The total retail square footage of retail space, both owned and leased;
o Pension expenses (where applicable);
o Executive management salary; and
o Governance risk indicators (i.e. GRId was recently launched as a new
measure of governance related risk which identifies and evaluates a
Select
an
industry
Pick companies
within time
frame
Research common reasons
for bankruptcy (general and
specific to firms in sample)
Collect
annual
reports
Collect
trade data
(TD)
Profile
industry
Extract financial
variables from Balance
Sheet (BS), Income
Statement (IS), Cash
Flow Statement (CF)
Extract managerial
decision-making
variables (MDM)
from Notes, MD&A,
Auditors Report
Data exploration (i.e. correlations)
Select set of
variables
Create metrics DEA
(IS, BS, CF, MDM
and/or TD)
Combine in
DEA Second
Stage
Evaluate against
benchmarks
(i.e. Altman)
Collect data on
economic and
market factors
pertinent to
industry
Reduce to
manageable
set
Compare average
performance to economic
and market trend
Worse?
Similar?
Different?
Modelling
complete
Opportunity to improve prediction
by incorporating economic and
market data into models
No Better?
Yes
168

companys key governance practices across four dimensions: Audit, Board of
Directors, Compensation and/or Remuneration, and shareholder rights. It
supports investors in their investment decision-making process making by
flagging firms with governance-related concerns) (ISS, 2012).
Using the trend of each variable (i.e. in the form of a slope or derivative for a linear
function) as a variable reflecting change over time.
Analyzing companies by further decomposition by business and geographical
segments, and discount or high-end brands.
Applying this methodology to firms in the private sector or at different stages in their
life cycle (i.e. start-up) where the risk of bankruptcy is higher.
Creating a technique that calibrates for time since data from companies do not
overlap entirely over the same period. This would increase the amount of useable
data.
Modelling with quarterly and/or monthly data and adjust for seasonality.
Examining further the similarities and differences between the layering and non-
layering approaches.
Conducting a more rigorous sensitivity analysis to test the robustness of the models
(i.e. testing the degree to which the overall second stage score is affected by the value
of each variable). In this thesis, sensitivity analysis was performed on each variable
to determine whether it should be included based on whether prediction improved.
Considering the cost or percentage loss by bankruptcy as the dependent or target
variable, as opposed to the percentage of companies that failed.
Determining and adjusting the models for the difference in cost of Type I error and
Type II error (as it was conservatively assumed equal in this thesis).
Resolving the limitations of zero and negative numbers for the SBM DEA model.
Incorporating non-discretionary and/or categorical variables (variables that one has
no control over, such as the market and economy) into DEA.
Investigating the inconsistency between normal and inverse DEA.

169

References
Aigner, D., Lovell, C.A.K., Schmidt, P., 1977. Formulation and estimation of stochastic
frontier production function models. Journal of Econometrics 6 (1), 21-37.

Albrecht, C., 2008. The nature of financial statement fraud. Internal Auditing 23 (4), 22-27.

Altman, E.I., 1968. Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy. Journal of Finance 23 (4), 589-609.

American Institute of CPAs, 2011. http://www.aicpa.org/Research/Standards/

Apostolou, B., Hassell, J.M., Webber, S.A. October 2001. Fraud risk factors: Rating by
forensic experts. CPA Journal, 48-52.

Banker, R.D., Charnes, A., Cooper W.W., 1984. Some models for estimating technical and
scale inefficiencies in data envelopment analysis. Management Science 30 (9), 1078-
1092.

Banker, R.D., Morey, R., 1986. Efficiency analysis for exogenously fixed inputs and
outputs. Operations Research 34 (4), 513-521.

Banker, R.D., Chang, H., Cooper, W.W., 1996. Simulation studies of efficiency, returns to
scale, and misspecification with nonlinear functions in DEA. Annals of Operations
Research 66, 233-253.

Barr, R. S., Seiford, L. M., Siems, T.F., 1993. An envelopment-analysis approach to
measuring the managerial efficiency of banks. Annals of Operations Research 45, 1
19.

Barr, R. S., Seiford, L.M., Siems, T.F., 1994. Forecasting bank failure: A non-parametric
frontier estimation approach. Recherches Economiques de Louvain 60 (4), 417429.

Bauer, P.W., Berger, A.N., Ferrier, G.D., Humphrey, D.B., 1998. Consistency conditions for
regulatory analysis of financial institutions: A comparison of frontier efficiency
methods. Journal of Economics and Business 50 (2), 85-114.

Beaver, W., 1966. Financial ratios as predictors of failure. Empirical Research in
Accounting: Selected Studies supplement to Journal of Accounting Research 4, 71-
111.

Bell, T.B., Carello, J.V., 2000. A decision aid for assessing the likelihood of fraudulent
financial reporting. Auditing: A Journal of Practice & Theory 169.

Beneish, M.D., 1999. The detection of earnings manipulation. Financial Analysts Journal 55
(5), 24-36.

170

Berger, A.N., 1993. Distribution-free estimates of efficiency in the U.S. banking industry
and tests of the standard distributional assumptions. Journal of Productivity Analysis 4
(3), 61-92.

Berger, A.N., Humphrey, D.B., 1997. Efficiency of financial institutions: International
survey and directions for future research. European Journal of Operational Research 8
(2), 175-212.

Blum, M., 1974. Failing company discriminant analysis. Journal of Accounting Research.

Bradley, D.B., Cowdery, C., 2004. Small business: Causes of bankruptcy. Internal
document of UCA Small Business Advancement National Center, 205-219.

Bureau of Economic Analysis, 2010. http://www.bea.gov/

Bureau of Labour Statistics, 2010. http://www.bls.gov/

Charnes, A., Cooper, W.W., Rhodes, E.L., 1978. Measuring the efficiency of decision
making units. European Journal of Operational Research 2 (6), 429-444.

Charnes, A., Cooper, W.W., Lewin, A.Y., Seiford, L.M., (Eds.), 1994. Data Envelopment
Analysis: Theory, Methodology and Applications, Kluwer Academic Publishers.

Cielen, A., Peeters, L., Vanhoof, K., 2004. Bankruptcy prediction using a data envelopment
analysis. European Journal of Operational Research 154 (2), 526-532.

Cummins, D., Zi, H., 1998. Comparison of frontier efficiency methods: An application to
the U.S. life insurance industry. Journal of Productivity Analysis 10.

Cooper, W.W., Seiford, L.M., Tone, K., 2007. Data Envelopment Analysis: A
Comprehensive Text with Models, Applications, References and DEA-Solver Software.
Springer, New York.

Deakin, E.B., 1972. A discriminant analysis of predictors of business failure. Journal of
Accounting Research.

Deloitte, 2008. 2008 Industry Outlook: A look around the corner.

Divine, J.D., 1986. Efficiency analysis and management of not for profit and governmentally
regulated organizations. Ph.D. Dissertation, Graduate School of Business, University
of Texas, Austin.

Edmister, R.O., 1972. An empirical test of financial ratio analysis for small business failure
prediction. Journal of Financial and Quantitative Analysis.

171

Elitzur, R., 2008. The accounting art of war: Enron and other cases. Advisor Guide to Risk
Management, The Bonham Centre for Finance, Joseph L. Rotman, School of
Management, 59-66.

Emrouznejad, A., Amin, G.R., 2007. DEA models for ratio data: Convexity consideration.
Applied Mathematical Modelling.

Epstein, B.J., Jermakowicz, E.K., 2007. Interpretation and Application of International
Financial Reporting Standards. John Wiley & Sons.

Farrell, M.J., 1957. The measurement of productive efficiency. Journal of Royal Statistical
Society, Series A 120 (3), 253-290.

Forrester Research, 2011. http://www.forrester.com/

Financial Accounting Standards Board, 1976. Accounting for leases. Statement of Financial
Accounting Standards No. 13.
Gill, J.O., 1994. Financial Basics of Small Business Success. Crisp Publications.
ISS, 2012. http://issgovernance.com/grid-info.
Hanson, R.O., 2003. A study of Altman's revised four-variable Z-score bankruptcy
prediction model as it applies to the service industry (Edward I. Altman). Dissertation
Abstracts International, 63-12A, 4375.

Hall, P., Simar, L., 2002. Estimating a changepoint boundary or frontier in presence of
observation error. Journal of the American Statistical Association 97, 523-534.

Hollingsworth, B., Smith, P.C., 2003. The use of ratios in data envelopment analysis.
Applied Economics Letters 10, 733-735.

International Council Shopping Centres, 2010. http://www.icsc.org/

Kaminski, K.A., Wetzel, T.S., Guan, L. 2004. Can financial ratios detect fraudulent
financial reporting? Managerial Auditing Journal 29 (1), 15-28.

Kao, C., Liu, S.T., 2004. Predicting bank performance with financial forecasts: A case of
Taiwan commercial banks. Journal of Banking & Finance 28, 23532368.

Kirkos, E., Charalambos, S., Manolopoulus, Y., 2007. Data mining technique for detection
of fraudulent financial statements. Expert Systems with Applications 32, 995-1003.

Land, K.C., Lovell, C.A., Thore, A., 1993. Chance-constrained data envelopment analysis.
Managerial and Decision Economics 14, 541-554.

172

Lewis, H.F., Sexton, T.R., 2004. Network DEA: efficiency analysis of organization with
complex internal structure. Computers & Operations Research 31, 1365-1410.

Libby, R., 1975. Accounting ratios and the prediction of failure: Some behavioural evidence.
Journal of Accounting Research.

Luenberger, D.G., 1998. Investment Science. Oxford University Press, NY.

National Retail Federation, 2010. http://www.nrf.com/

Moyer, R., 1977. Forecasting financial failure: a re-examination. Financial Management.

Ohlson, J.A., 1980. Financial ratios and the probabilistic prediction of bankruptcy. Journal
of Accounting Research 18 (1), 109-131.

zpeynirci, ., Kksalan, M., 2007. Performance evaluation using data envelopment
analysis in the presence of time lags. Journal of Productivity Analysis 27, 221-229.

Paradi, J.C., Asmild, M., Simak, P.C., 2004. Using DEA and worst practice DEA in credit
risk evaluation. Journal of Productivity Analysis 21, 153-165.

Paradi, J.C, Rouatt, S.M., Zhu, H., 2011. Two stage evaluation of bank branch efficiency
using data envelopment analysis. Omega 29, 99-109.

Punik, K., Tajnikar, M., 2008. Technical and cost efficiencies as determinants of business
failures of small firms. Eastern European Economics 46 (1), 43-62.

Premachandra, M., Chen, Y., Watson, J., 2011. DEA as a tool for predicting corporate
failure and success: A case of bankruptcy assessment. Omega 39 (6), 620-626.

Ramamoorti, S., Olsen, W., 2007. Fraud: the human factor. Financial Executive, 53-55.

Ravi Kumar, P., Ravi, V., 2006. Bankruptcy prediction in banks and firms via statistical and
intelligent techniques a review. European Journal of Operational Research 180, 1-28.

Schmidt, P., Sickles, R.C., 1984. Production frontiers and panel data. Journal of Business
and Economic Statistics 2 (4), 221-250.

Sengupta, J.K., 1996. Dynamic data envelopment analysis. International Journal of Systems
Science 27 (3), 277-284.

Seow, J., 2009. Cue usages in financial statement fraud risk assessments: effects of
technical knowledge and decision aid uses. Accounting and Finance 49, 183-205.

Smith, M., Baharuddin, I., 2005. Auditors perception of fraud risk indicators. Managerial
Auditing Journal 20 (1), 73-85.
173

Shetty, U., Pakkala, T.P.M., Mallikarjunappa, T., 2012. A modified directional distance
formulation of DEA to assess bankruptcy: An application to IT/ITES companies in
India. Expert Systems with Applications 39, 1988-1997.

Sueyoshi, T., 1998. DEA-discrimant analysis in the view of goal programming. European
Journal of Operational Research 22, 349-376.

Sueyoshi., T., Goto, M., 2009. Methodological comparison between DEA (data envelopment
analysis) and DEA-DA (discriminant analysis) from the perspective of bankruptcy
assessment. European Journal of Operational Research 199, 561-575.

Stickney, C.P., Brown, P., Wahlen, J.M., 2006. Financial Reporting, Financial Statement
Analysis, and Valuation: A Strategic Perspective. 6
th
Edition, South-Western College
Pub.

Thanassoulis, E., 1999. Setting achievement targets for school children. Education
Economics 7 (2), 101-119.

Thompson, R.G., Singleton, F.D., Thrall, R.M., Smith, B.A., 1986. Comparative site
evaluations for locating a high-energy physics lab in Texas. Interfaces 16, 35-49.

Tone, K., 2001. A slacks-based measure of efficiency in data envelopment analysis.
European Journal of Operational Research 130 (3), 498-509.

Tran, A., 2007. Two-stage financial risk tolerance assessment using data envelopment
analysis. Masters Dissertation, University of Toronto.

U.S. Census Bureau, 2010. http://www.census.gov/econ/industry/current/c448150.htm

U.S. Courts, Bankruptcy Statistics, 2010.
http://www.uscourts.gov/bnkrpctystats/bankruptcystats.htm

U.S. Department of Commerce, 2010. http://www.commerce.gov

U.S. Department of Labour, 2010. http://www.dol.gov

U.S. Municipal Bond Fairness Act, 2008.

Weidman, C.I., 1999. Instructional case: Detecting earnings manipulation. Issues in
Accounting Education 14 (1), 145-176.

Wilcox, J.W., 1971. A simple theory of financial ratios as predictors of failure. Journal of
Accounting Research 9 (2), 389-395.

Wilks, T.J., Zimbelman, M.F., 2004. Using game theory and strategic reasoning concepts to
prevent and detect fraud. Accounting Horizons 18 (3), 173-184.
174

Wongphatarakul, V., Friendlander, S.K., Pinto, J.P., 1998. A comparative study of PM
2.5

ambient aerosol chemical databases. Environmental Science Technology 32, 3926-
3934.

Xu, X., Wang, Y., 2009. Financial failure prediction using efficiency as a predictor. Expert
Systems with Applications 36, 366-373.

175

Appendix A Parametric Frontier Methodologies
There are three major parametric frontier efficiency methodologies: Stochastic Frontier
Analysis (SFA), the Thick-Frontier Approach (TFA) and the Distribution-Free Approach
(DFA). Unlike DEA, parametric methods must satisfy both technical and allocative
efficiency. They require the specification of the shape of the frontier and functional form,
limited to single output multiple input, or single input multiple output cases. However,
they are less likely to misidentify measurement error, transitory differences in cost, or
random error as inefficiency (Berger and Humphrey, 1997). SFA, TFA and DFA differ in
the distributional assumptions imposed to separate random error from efficiency.

A.1 Stochastic Frontier Analysis (SFA)
Stochastic Frontier Analysis (SFA) was introduced by Aigner, Lovell and Schmidt (1977). It
is a model in which inefficiencies are always bounded and nonnegative, and assumed to
follow an asymmetric distribution (i.e. half-normal), while random errors are assumed to
follow a symmetric distribution (i.e. standard normal). Inefficiencies and errors are
assumed orthogonal to the input, output and environmental variables specified in the cost
function = + . The estimated inefficiency for any DMU is taken as the conditional mean
of the distribution of given the estimate of (Berger and Humphrey, 1997). While SFA is
criticized for the somewhat arbitrary distributional assumptions, a positive aspect is that it
ranks the efficiencies of the DMUs in the same order as their cost function residuals, no
matter what assumptions are imposed (Bauer et al., 1998).

A.2 Thick Frontier Approach (TFA)
Thick Frontier Approach (TFA) employs the same form for the frontier cost function as SFA
but is based on a regression that measures a general level of overall efficiency for all DMUs,
rather than point estimates for individual DMUs to reduce the effect of extreme points in
the sample. This general level of overall efficiency is the estimate of efficiency differences
between the best and worst quartile. The thick frontier of efficient DMUs is calculated from
data limited to the lowest-cost quartile representing random error only. Like SFA, TFAs
rather arbitrary assumptions with respect to the thickness of the frontier are questionable, but
still has appeal with rank orderings (Berger and Humphrey, 1997).
176

A.3 Distribution-Free Approach (DFA)
Similar to SFA and TFA, the Distribution-Free Approach (DFA), developed by Schmidt and
Sickles (1984) and Berger (1993), specifies a functional form for the cost function but
provides an alternative procedure for estimating relative firm inefficiency. In contrast, it
does not specify the distribution of efficiency nor impose that deviations within one group
are all random errors and deviations between groups are all inefficiencies. Instead, DFA
assumes that each DMU has a core efficiency that is constant over time and random error
averages out to zero over time. Nonetheless, DFA is criticized for having somewhat arbitrary
assumptions, such as inefficiency being the only time-invariant fixed effect (Bauer et al.,
1998).

177

Appendix B Examples of Data Envelopment Analysis
For simplicity, these graphical examples consider only one or two inputs and outputs. Unless
otherwise stated, CCR efficiency is calculated in these examples.

B.1 One Input, One Output
Consider 4 DMUs (A, B, C and D), each with one input and one output (Figure B.1). A, B
and C are BCC-efficient but only B is CCR-efficient.

Figure B.1: One Input and One Output Case
If the model is input-oriented, the BCC efficiency of D is:
667 . 0
4
7 . 2
*
,
= = =
PD
PR
BCC D

while its CCR efficiency is:
5625 . 0
4
25 . 2
*
,
= = =
PD
PQ
CCR D

Therefore, BBC and CCR efficiency can be achieved by reducing the existing input value (4)
by 332 . 1 ) 667 . 0 1 ( 4 = units and 75 . 1 ) 5625 . 0 1 ( 4 = units, respectively.
If the model is output-oriented, the BCC efficiency of D is:
667 . 1
*
,
= =
DT
ST
BCC D

while its CCR efficiency is:
778 . 1
3
333 . 5
*
,
= = =
DT
UT
CCR D

0
1
2
3
4
5
6
7
0 1 2 3 4 5 6
A
B
C
D
CCR Frontier
BCC Frontier
X
X X
S
R
Q
X
P
X
T
X
U
Input
Output
0
1
2
3
4
5
6
7
0 1 2 3 4 5 6
A
B
C
D
CCR Frontier
BCC Frontier
X
X X
S
R
Q
X
P
X
T
X
U
Input
Output
178

In other words, achievement of efficiency requires augmenting Ds output from its observed
value (3) to the projected value on the frontier (5) by 2 ) 1 667 . 1 ( 3 = units. Note that
*
,
*
,
1
CCR D
CCR D
= , but this simple reciprocal relation between input- and output- oriented
models is not applicable to the BCC model (Cooper et al., 2005).

B.2 Minimize Two Inputs, Unitized Output
Suppose there are 7 DMUs (A to G) which use two inputs to produce one output (Table B.1,
Figure B.2). Note that the output is unitized.
Table B.1: Two Inputs and One Output Case
DMU A B C D E F G
Input
x
1
4 7 8 4 2 10 3
x
2
3 3 1 2 4 1 7
Output y 1 1 1 1 1 1 1

Figure B.2: Two Inputs and One Output Case

CCR efficiency of A can be calculated by two methods. A line can be projected
radially from the origin to A, crossing through the frontier at P (the projection point).
Efficiency is then the ratio of the distance from the origin to P, to the distance from the origin
to A:
8571 . 0
*
= =
OA
OP
A

Therefore, A can be improved by reducing its inputs to 0.8571 of their existing values; i.e. P
) 6 . 2 , 4 . 3 (
2 1
x x .
Efficiency can also be determined with the following linear program:
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7 8 9 10 11
A
G
B
C
D
E
F
X
X
Q
P
O
x
2
/y
x
1
/y
x
2
/y
x
1
/y
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7 8 9 10 11
A
G
B
C
D
E
F
X
X
Q
P
O
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7 8 9 10 11
A
G
B
C
D
E
F
X
X
Q
P
O
x
2
/y
x
1
/y
x
2
/y
x
1
/y
179

1
0 7 4 2 3 3 3
0 3 10 2 4 8 7 4 4
To Subject
Minimize : II Phase
Minimize : I Phase
2
1
2 1
= + + + + + +
=
=

+
+
s
s
s
s s s
G F E D C B A
G F E D C B A A
G F E D C B A A
A

which yields an optimal solution of 8571 . 0
*
=
A
with 7143 . 0
*
=
D
, 2857 . 0
*
=
E
,
0
* * * * *
= = = = =
G F C B A
and 0
* *
2
*
1
= = =
+
s s s . The reference set for A is E
A
=
{D, E}.
*
D
and
*
E
are the proportions contributed by D and E to the point on the frontier P
used to evaluate A. P is also a hypothetical DMU which is made from 0.7143 of Ds inputs
and 0.2857 of Es inputs. Its coordinates ) , (
2 1
x x can be determined as follows:
5714 . 2 4 2857 . 0 2 7143 . 0 2857 . 0 7143 . 0
4286 . 3 2 2857 . 0 4 7143 . 0 2857 . 0 7143 . 0
2 2 2
1 1 1
+ = + =
+ = + =
E D
E D
x x x
x x x

Alternatively, this can be interpreted in terms of CCR projection, which is achieved by a
14.29% reduction in both inputs.
1
5714 . 2 0 4 8571 . 0
4286 . 3 0 4 8571 . 0
*
1
2 2
*
2
1 1
*
1
= +
= =
= =
+
s y y
s x x
s x x
-*
A
-*
A

For the case where there are slacks (such as G), the following LP,
1
0 7 4 2 3 3 7
0 3 10 2 4 8 7 4 3
To Subject
Minimize : II Phase
Minimize : I Phase
2
1
2 1
= + + + + + +
=
=

+
+
s
s
s
s s s
G F E D C B A
G F E D C B A A
G F E D C B A A
G

yields 6667 . 0
*
=
G
with 1
*
=
E
, 0
* * * * * *
= = = = = =
G F D C B A
and 0
*
1
=
s ,
6667 . 0
*
2
=
s , 0
*
=
+
s . The reference set for G is E
G
= {E}. Because of excess in input 2 (
6667 . 0
*
2
=
s ), CCR projection for G is achieved by a 33.33% reduction in input 1 and a

42.86% reduction in input 2 (Cooper et al., 2007).
180

1
4 6667 . 0 4 6667 . 0
2 0 3 6667 . 0
*
1
2 2
*
2
1 1
*
1
= +
= =
= =
+
s y y
s x x
s x x
-*
G
-*
G

B.3 Maximize Two Outputs, Unitized Input
Consider 7 DMUs (A to G), each consuming one input to produce two outputs (Table B.2,
Figure B.3). Note that the input is unitized.
Table B.2: One Input and Two Outputs Case
DMU A B C D E F G
Input x 1 1 1 1 1 1 1
Output
y
1
1 2 3 4 4 5 6
y
2
5 7 4 3 6 5 2

Figure B.3: One Input and Two Outputs Case

CCR efficiency of D can be calculated by two methods. A line can be projected
radially from the origin through D and onto the frontier at P (the projection point).
Efficiency is then the ratio of the distance from the origin to D, to the distance from the
origin to P the hypothetical DMU that D aims to be as efficient as.
= =
OD
OP
D
*
1.333
For D to be efficient, all of its outputs must be increased by 1.333 times their current values
(4, 3); that is, D must project to P ) 4 , 333 . 5 ( ) 3 , 4 ( 333 . 1 ) , (
2 1
= = y y .
Alternatively, the following LP would yield the same results:
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
X
X
B
A
D
G
F
E
C
Q
P
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
X
X
B
A
D
G
F
E
C
Q
P
y
2
/x
y
1
/x
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
X
X
B
A
D
G
F
E
C
Q
P
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
X
X
B
A
D
G
F
E
C
Q
P
0
1
2
3
4
5
6
7
8
0 1 2 3 4 5 6 7
X
X
B
A
D
G
F
E
C
Q
P
y
2
/x
y
1
/x
181

1
0 2 5 6 3 4 7 5 2
0 6 5 4 4 3 2 1 6
To Subject
Maximize : II Phase
e Maximiz : I Phase
2
1
2 1
= + + + + + + +
= +
= +
+ +
+
+
s
s
s
s s s
G F E D C B A
G F E D C B A G
G F E D C B A G
G

with the projection:
-*
B
B
t x x
t y y
t y y

+
+
+
+
1
*
2 2
*
2
*
1 1
*
1

For DMUs with output slacks (such as A), efficiency is calculated in two phases: first,
a radial projection to Q on the frontier followed by an expansion to B. For example, A has a
technical efficiency of 41 . 1
*
= =
OA
OQ
A
but note that after its projection to Q, it is still not as
efficient as B. Hence, for A to be fully efficient, in addition to increasing all its outputs by
1.4 times their current value, output 1 must be increased from 1.4 to 2 (Cooper et al., 2007).

182

Appendix C Reasons for Bankruptcy as Cited by Press Releases
Filenes Basement Corp (BSMT), Bankruptcy in 1999
- Chapter 11 filing prompted by low inventory caused by vendors refusing to deliver
merchandise, as well as competition (abundance of off-pricers in Northeast)
- Embarked on an ill-advised strategy of expansion in 1988 when acquired by May
Department Stores, attempted private label strategy
- High operating losses
- Second filing in 2009: recession and consumers pulling back made debt burden
unmanageable

Christopher & Banks Corp. (CBK), Bankruptcy in 1996
- Chapter 11 filed after experiencing three-year retail slowdown and consecutive
annual losses, as well as debt problems it had been battling since 1986
- Doing well since restructuring and renamed from Brauns Fashion Corporation in July
2000

Casual Male Corp (CMALQ), Bankruptcy in 2001
- Chapter 11 filing stemmed from a need to restructure its long-term indebtedness
- Delisted by NASDAQ as stock did not closed above the minimum $5 per share
required

CML Group (CML), Acquired and Bankruptcy in 1998
- Chapter 11 filing did not surprise industry observers given that the company had not
reported an annual profit since 1994

County Seat Stores, Inc. (CSS), Bankruptcy in 1997
- Severe liquidity problems
- Strategy over last two years produced bloated inventories and labels that did not yield
enough profit margin to keep it afloat

Clothestime, Inc. (CTMEQ), Bankruptcy in 1996
- During the beginning of 1996, sales dropped 36.2% to $103.2M and net losses grew
to from $10M to $13.8M.
- Too much owed to landlords, banks and other creditors
- Pummeling stock which traded two years ago at above $10 to $1.25 in NASDAQ
- Competition: customers flock to Wal-Mart, Sears and JC Penney that can offer
familiar name brands at attractive prices; this is deadly for smaller specialty chains
like Clothestime which were profitable by selling their own private-label apparel at
low prices

Eddie Bauer Holdings Inc. (EBHI), Bankruptcy in 2009
- With recession, consumers curtailed spending and the company struggled to pay its
debt
- Falloff in sales came as company was trying to pull off a multiyear turnaround that
included cost cuts and changes to its management team and merchandise

183

Edison Brothers Stores, Inc. (EDBRW), Bankruptcy in 1996
- Cited severe competition and disappointing operating results (losses) for filing
Chapter 11
- Other reasons included a poor specialty retail environment and bleak holiday season
outlook

Just for Feet Inc. (FEETQ), Bankruptcy in 2000
- Collapsed in 1999 amid an accounting fraud: 3 former executives pleaded guilty to
crimes related to overstating earnings by $8M between 1996 and 1998
- When it filed, company cited lagging sales, excess inventories and high operating
costs for losses of more than $21M in the six months ended July 21, 1999

Footstar Inc. (FTAR), Bankruptcy in 2002
- SEC found discrepancies in accounts payable balances, namely an understatement of
$35M in its athletics segment that forced company to restate its financials from 1997
through 2002 which reduced earnings by as much as $53M over the five-and-a-half-
year period

Factory-2-U Stores, Inc (FTUS), Bankruptcy in 2004
- Filed Chapter 11 to restructure $73.5M in debt after struggling through a weak
holiday shopping season (soft sales) and competition from bigger rivals such as Wal-
Mart, Target, Kmart
- Management was in chaos with CEO and CFO resignations

Gadzooks, Inc. (GADZQ), Bankruptcy in 2005
- Heavy debt load prompted reorganization of core business around 252 stores chosen
to strengthen its market position, and develop a plan to boost sales and profit

Jay Jacobs (JAYJ), Bankruptcy in 1994
- Consequence of financial slide was company losing trade support of its vendors,
making it difficult for the stores to get merchandise
- Filed Chapter 11 to get out of leases at unprofitable store sites

Gantos / Kinder Holding Corp. (KDRH), Bankruptcy in 1993
- Pressed by falling sales, big losses, growing debt, and plummeting stock
- Tailspin caused by slow down in retail apparel market and ill-conceived
merchandising rollouts

Harolds Stores (HRLS), Bankruptcy in 2008
- Ceased operations because of increased competition and weak sales

Kenwin Shops, Inc. (KNWN), Bankruptcy in 1994
- General economic decline eroded financial position (net losses) and forced several
vendors to stop shipping
184

- Quality and attractiveness of merchandise did not satisfy consumers' desires causing
company to lose customers and sales. Also, overhead and selling expenses were high
and company operated numerous stores which were unprofitable

Lamonts Apparel, Inc. (LAM), Bankruptcy in 1995
- Filed Chapter 11 to get out of costly store leases and to reduce insolvency
(haemorrhage of cash related to its debt service)
- As Lamonts expand, its financial performance deteriorated with general downturn in
retail industry and competition from discount retailers like Target and Wal-Mart
- Critics claimed Lamonts failed to keep pace with market changes and was operating
with an obsolete strategy, with net losses $8.5M, $11.7M, $19.3M and $10.8M from
1990-1993 despite increase in sales

Loehmanns Inc. (LOEHQ), Bankruptcy in 1999
- Declining sales and an inability to carry debt load
- More specifically: failure to make a $5.5M interest payment to creditors; lost more
than $5M last year on sales of $432M; CFO resigned abruptly without explanation;
Moody's Investors Service Inc. downgraded its debt to a status lower than junk;
many vendors stopped shipping to company; competition from discount chains

Merry-Go-Round Enterprises Inc. (MGR), Bankruptcy in 1994
- Filed Chapter 11 after a long but futile effort to placate its creditors (debt)
- Fell into a cash-flow crisis by overestimating the appeal of "hip-hop" clothing which
forced the company to take a $35.1M charge from having to discount heavily to sell.

One Price Clothing Stores, Inc. (ONPRQ), Bankruptcy in 2004
- Plagued by liquidity issues

Paul Harris Stores, Inc. (PHRS), Bankruptcy in 2000
- Plagued by liquidity problems and slow sales
- Problems developed after company cut prices deeply to sell off inventory earlier in
the year, hurting results and cash squeeze made it difficult to get new merchandise

Shoe Pavilion, Inc. (SHOE), Bankruptcy in 2008
- Five straight quarterly losses due to recession

Todays Man, Inc. (TMAN), Bankruptcy in 1996 and 2002
- Company blamed bad economy, skittish suppliers unwilling to extend credit, a trend
among men toward more casual dress, consumers spending more money on
computers than clothing, going-out-of-business sales that cut into market share, and
merchants so desperate to bring customers into their stores that they practically gave
away the goods at extreme discounts (competition)
- Company had too rapid growth, was unprofitable, defaulted on loan and had credit
line reduced

185

Appendix D Supplementary Background
D.1 Credit Rating Agencies
Table D.1: Broad Categories for Economic Indicators
Category Description Examples
Total Output, Income
and Spending
Broad measures of economic
performance
GDP, National Income, Consumption Expenditure,
Corporate Profits
Employment,
Unemployment and
Wages
Measures of the labour
market
Unemployment Rate, Average Weekly Hours and Earnings,
Labour Productivity
Production and
Business Activity
Measures of how much
businesses are producing and
the level of new construction
Industrial Production and Capacity Utilization, New
Construction, New Private Housing and Vacancy Rates,
Business Sales and Inventories, Manufacturers Shipments,
Inventories and Orders
Prices Measures of what consumers
and businesses pay for
commodities
Producer Prices, Consumer Prices, Prices Received and
Paid By Farmers
Money, Credit and
Security Markets
Measures of the amount of
money in the economy and
interest rates
Money Stock (M1, M2 and M3), Bank Credit at All
Commercial Banks, Consumer Credit, Interest Rates and
Bond Yields, Stock Prices
Federal Finance Measures of government
spending, deficits and costs
Federal Receipts (Revenue), Federal Outlays (Expenses),
Federal Debt
International Statistics Measures of how much the
country is exporting and
importing
Industrial Production and Consumer Prices of Major
Industrial Countries, U.S. International Trade in Goods and
Services, U.S. International Transactions

Table D.2: Rating Codes for Each Credit Rating Agencies
Moodys S&P Fitch
Prime Aaa AAA AAA
High Grade
Aa1 AA+ AA+
Aa2 AA AA
Aa3 AA- AA-
Upper Medium Grade
A1 A+ A+
A2 A A
A3 A- A-
Lower Medium Grade
Baa1 BBB+ BBB+
Baa2 BBB BBB
Baa3 BBB- BBB-
Non-Investment Grade Speculative
Ba1 BB+ BB+
Ba2 BB BB
Ba3 BB- BB-
Highly Speculative
B1 B+ B+
B2 B B
B3 B- B-
Substantial Risk Caa1 CCC+
CCC
Extremely Speculative Caa2 CCC
In default with little prospect for recovery
Caa3 CCC-
Ca
CC
C
In default
C
D
DDD
/
DD
D
186

Appendix E List of Metrics
Table E.1: Variables from Principal Component Analysis and Logistic Regression
# Inputs Outputs
1
AD, AR,CA, CL, COGS, Inv, Other
NonCA, PPE, SGA, TA, TL
Rev, SE
2 AP, CM, IE, PPE, SGA, TA, TL Cash, NI, Rev, SE

Table E.2: Profitability and Success Metrics (Normal DEA Frontier)
# Inputs Outputs
1 AR, COGS, IE, Inv, PPE, SGA, TA, TL NI, Rev
2 AR, COGS, IE, Inv, PPE, SGA, TA, TL NI, RE, Rev
3 AR, COGS, IE, Inv, PPE, SGA, TA, TL NI, Rev, SE
4 AR, COGS, IE, Inv, PPE, SGA, TA, TL Rev
5 AR, COGS, IE, Inv, PPE, SGA, TA, TL RE, Rev
6 AR, COGS, IE, Inv, PPE, SGA, TA, TL Rev, SE
7 AR, IE, Inv, PPE, SGA, TA, TL Rev, SE
8 COGS, IE, Inv, PPE, SGA, TA, TL CFO, Rev, SE
9 COGS, IE, Inv, PPE, SGA, TA, TL CFO, RE, Rev, SE
10 COGS, IE, Inv, PPE, SGA, TA, TL NI, Rev, RE, SE
11 COGS, IE, Inv, PPE, SGA, TA, TL RE, Rev, SE
12 COGS, IE, Inv, PPE, SGA, TA, TL Rev, SE
13 COGS, SGA, TA, TL NI, Rev, SE
14 COGS, SGA, TA, TL NI, Rev, SE
15 COGS, SGA, TA, TL NI, Rev, RE, SE
16 IE, TA CFO, EBIT, RE, WC
17 IE, TA CFO, NI, RE, Rev
18 IE, TA NI, RE, Rev
19 IE, TA NI, RE, Rev, SE
20 IE, TA, TL CFO, NI, RE, Rev
21 IE, TA, TL NI, RE, Rev
22 IE, TA, TL NI, RE, Rev, SE
23 IE, TA, TL Rev, SE
24 SGA, PPE, TA, TL Rev, SE
25 TA CFO, EBIT, RE
26 TA CFO, NI, RE
27 TA CFO, NI, RE, Rev
28 TA, TL CFO, NI, RE, Rev
29 TA, TL EBIT, RE, SE,WC
30 TA, TL NI, RE, Rev
31 TA, TL NI, RE, Rev, SE
32 TA, TL Rev, SE

Table E.3: Profitability and Success Metrics (Inverse DEA Frontier)
# Inputs Outputs
1 CFO, NI, RE, Rev TA, TL, IE
2 NI, RE, Rev IE, TA
3 NI, RE, Rev IE, TA, TL
4 Rev, SE AR, COGS, IE, Inv, PPE, SGA, TA, TL
5 Rev, SE COGS, IE, Inv, PPE, SGA, TA, TL
6 Rev, SE IE, TA, TL
7 Rev, SE SGA, PPE, TA, TL

187

Table E.4: Bankruptcy and Failure Metrics (Worst-Practice DEA Frontier)

# Inputs Outputs
1 AR, Cash, MS, NR CL
2 AR, Cash, Lev, MS, NR CL
3 CA, CFO CL
4 CA, CFO CL, NI
5 CA, CFO, EBIT, RE CL, IE,TA
6 CA, CFO, Lev CL
7 CFO, EBIT IE
8 CFO, EBIT, Lev IE
9 CFO, EBIT, Lev, SE, TA IE, TL
10 CFO, EBIT, SE, TA IE, TL
11 CFO, TA, WC IE
12 CFO, WC IE
13 Lev, SE, TA TL
14 SE, TA TL

Table E.5: Bankruptcy and Failure Metrics (Bad Outputs Model)
# Inputs Outputs
1 AR, COGS, IE, Inv, PPE, SGA, TA, TL Rev, SE
4 IE, TA NI, RE, Rev
5 IE, TA, TL CFO, NI, RE, Rev
6 IE, TA, TL NI, RE, Rev
7 IE, TA, TL Rev, SE
8 PPE , SGA,TA, TL Rev, SE

Table E.6: Balance Sheet Metrics (Normal DEA Frontier)
# Inputs Outputs
1
AR, CA, Goodwill, Inv, MS, Net PPE,
LTIS, TA
Cash, RE, SE
2 AP, CL, CM, LTD, NPSTD, TL RE, SE
3
AR, AP, CA, CL, CM, Goodwill, Inv,
MS, Net PPE, NPSTD, LTD, LTIS, TA,
TL
Cash, RE, SE

Table E.7: Income Statement Metrics (Normal DEA Frontier)
# Inputs Outputs
1 COGS, ITE, Net IE, SGA NI, Rev

Table E.8: Qualitative Metrics (Normal DEA Frontier)
# Inputs Outputs
1
Auditors Opinion, Auditor Turnover,
Legal Proceedings, Management
Turnover, Related Parties, Retirement
Plans
Dummy
31

31
188

Table E.9: Variables for Mega Model
Source Variable Orientation
Income Statement
COGS Input
ITE Input
Net IE Input (+) / Output (-)
NI Output (+) / Input (-)
Rev Output
SGA Input
Balance Sheet
AR Input
AP Input
CA Input
Cash Output
CL Input
CM Input
Goodwill Input
Inv Input
LTD Input
LTIS Input
MS Input
Net PPE Input
NPSTD Input
RE Output (+) / Input (-)
SE Output (+) / Input (-)
TA Input
TL Input
Cash Flow Statement
CFO Output (+) / Input (-)
CFI Output (+) / Input (-)
CFF Input (+) / Output (-)
Managerial Decision-Making
Auditors Opinion Input
Auditor Turnover Input
Legal Proceedings Input
Management Turnover Input
Related Parties Input
Retirement Plans Input
Market Factors
DJIA Non-Discretionary Output
S&P 500 Non-Discretionary Output
NASDAQ-100 Non-Discretionary Output
Economic Factors
GDP Growth rate Non-Discretionary Output
Inflation Non-Discretionary Output
Interest Rate Non-Discretionary Output
Unemployment Rate Non-Discretionary Input
Debt as a Percentage of GDP Non-Discretionary Input
Personal Consumption Expenditures: Clothing Non-Discretionary Output
GDP: Clothing Non-Discretionary Output
Apparel Labour Productivity Non-Discretionary Output
Apparel Imports Non-Discretionary Output
Median Months for Sale (Housing) Non-Discretionary Input
Housing Units Started Non-Discretionary Output
Cotton Price Non-Discretionary Input
Oil Price Non-Discretionary Input

189

Appendix F Modelling Results
F.1 Logistic Regression (LR) Metric
Table F.1: Normal Scores Calculated with Variables Derived from LR by State
All A 649 0.38 0.36 0.26 0 1
All B 50 0.17 0.30 0.00 0 1
B 13 0.16 0.37 0.00 0 1
B-1 12 0.06 0.11 0.00 0 0.35
B-2 13 0.23 0.37 0.03 0 1
B-3 12 0.22 0.28 0.13 0 1
Total 699 0.37 0.36 0.24 0 1

Table F.2: Confusion Matrix for LR Normal Model

Actual

Active Bankrupt
Predicted
Active 157 5
Bankrupt 388 41

Figure F.1: Distribution of Scores Generated from LR Inverse DEA Model
Table F.3: Inverse Scores Calculated with Variables Derived from LR by State
All A 651 0.16 0.36 0.00 0 1
All B 50 0.43 0.49 0.01 0 1
B 13 0.48 0.50 0.25 0 1
B-1 12 0.50 0.52 0.51 0 1
B-2 13 0.39 0.51 0.00 0 1
B-3 12 0.33 0.49 0.00 0 1
Total 701 0.18 0.38 0.00 0 1
0
100
200
300
400
500
600
0
0
.
0
5
0
.
1
0
.
1
5
0
.
2
0
.
2
5
0
.
3
0
.
3
5
0
.
4
0
.
4
5
0
.
5
0
.
5
5
0
.
6
0
.
6
5
0
.
7
0
.
7
5
0
.
8
0
.
8
5
0
.
9
0
.
9
51
F
r
e
q
u
e
n
c
y
Bin for DEA Scores
Bankrupt
Active
190

Table F.4: Confusion Matrix for LR Inverse Model

Actual

Active Bankrupt
Predicted
Active 548 29
Bankrupt 103 21

F.2 Income Statement (IS) Metric

Figure F.2: Prediction by IS Metric from Two Years Back

Figure F.3: Prediction by IS Metric from Three Years Back

0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
191

F.3 Balance Sheet Assets (BSA) Metric

Figure F.4: Prediction by BSA Metric from Two Years Back

Figure F.5: Prediction by BSA Metric from Three Years Back

0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8
Layer
Type I Error
Type II Error
192

F.4 Balance Sheet Liabilities (BSL) Metric

Figure F.6: Prediction by BSL Metric from Two Years Back

Figure F.7: Prediction by BSL Metric from Three Years Back

0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
Layer
Type I Error
Type II Error
193

F.5 Managerial Decision-Making (MDM) Metric

Figure F.8: Prediction by MDM Metric from Two Years Back

Figure F.9: Prediction by MDM Metric from Three Years Back

0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6
Layer
Type I Error
Type II Error
194

F.6 Second Stage Models

Figure F.10: Classification by Layer in Absolute Numbers from Two Years Back

Figure F.11: Classification by Layer in Percentages from Two Years Back

Figure F.12: Prediction by Second Stage Model from Two Years Back
0%
20%
40%
60%
80%
100%
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
on Frontier
# DMUs on
Frontier
Layer
0%
20%
40%
60%
80%
100%
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
on Frontier
% of DMU on
Frontier
Layer
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Layer
Type I Error
Type II Error
195

Figure F.13: Classification by Layer in Absolute Numbers Three Years Back

Figure F.14: Classification by Layer in Percentages from Three Years Back

Figure F.15: Prediction by Second Stage Model from Three Years Back

0%
20%
40%
60%
80%
100%
0
10
20
30
40
50
60
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
on Frontier
# DMUs on
Frontier
Layer
0%
20%
40%
60%
80%
100%
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
on Frontier
% of DMU on
Frontier
Layer
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Layer
Type I Error
Type II Error
196

F.7 Projections for Active Companies
Figure F.16: IS Projections for COGS
Mean = 1.2% 2.7%; Median = 0%
Figure F.17: IS Projections for SGA
Mean = 1.9% 6.1%; Median = 0%
Figure F.18: IS Projections for Net Interest Expense
Mean = 23% 36%; Median = 0%
Figure F.19: IS Projections for Income Tax Expense
Mean = 12% 27%; Median = 0%
Figure F.20: IS Projections for Revenue
Mean = 0.3% 0.1%; Median = 0%
Figure F.21: IS Projections for Net Income
Mean = 260% 360%; Median = 73%
0
50
100
150
200
250
300
350
400
450
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
300
350
400
450
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
300
350
400
450
500
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
600
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
0
25
50
75
100
125
150
175
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
197

Figure F.22: BSA Projections for Marketable
Securities
Mean = 24% 41%; Median = 0%

Figure F.23: BSA Projections for Accounts
Receivable
Mean = 66% 44%; Median = 98%

Figure F.24: BSA Projections for Inventories
Mean = 39% 29%; Median = 44%

Figure F.25: BSA Projections for Current Assets
Mean = 3.9% 9.5%; Median = 0%

Figure F.26: BSA Projections for Long-Term
Investments in Securities
Mean = 8% 26%; Median = 0%

Figure F.27: BSA Projections for Net PPE
Mean = 21% 26%; Median = 5%
0
100
200
300
400
500
F
r
e
q
u
e
n
c
y
0
100
200
300
400
F
r
e
q
u
e
n
c
y
0
25
50
75
100
125
150
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
600
F
r
e
q
u
e
n
c
y
0
100
200
300
F
r
e
q
u
e
n
c
y
198

Figure F.28: BSA Projections for Goodwill
Mean = 30% 43%; Median = 0%

Figure F.29: BSA Projections for Total Assets
Mean = 12% 12%; Median = 10%

Figure F.30: BSA Projections for Cash
Mean = 370% 390%; Median = 196%

Figure F.31: BSA Projections for Retained Earnings
Mean = 247% 350%; ; Median = 79%

Figure F.32: BSA Projections for Shareholders
Equity
Mean = 28% 130%; Median = 0%

Figure F.33: BSL Projections for Accounts Payable
Mean = 28% 26%; Median = 23%

0
50
100
150
200
250
300
350
400
450
F
r
e
q
u
e
n
c
y
0
50
100
150
200
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
0
25
50
75
100
125
150
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
200
250
300
350
400
450
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
200
F
r
e
q
u
e
n
c
y
199

Figure F.34: BSL Projections for Notes Payable and
Short-Term Debt
Mean = 15% 35%; Median = 0%

Figure F.35: BSL Projections for Current Maturities
of Long-Term Debt
Mean = 37% 31%; Median = 40%

Figure F.36: BSL Projections for Current Liabilities
Mean = 28% 24%; Median = 23%

Figure F.37: BSL Projections for Long-Term Debt
Mean = 54% 49%; Median = 93%

Figure F.38: BSL Projections for Total Liabilities
Mean = 27% 28%; Median = 18%

Figure F.39: BSL Projections for Retained Earnings
Mean = 370% 390%; Median = 200%
0
100
200
300
400
500
600
F
r
e
q
u
e
n
c
y
0
25
50
75
100
125
150
175
200
F
r
e
q
u
e
n
c
y
0
25
50
75
100
125
150
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
300
350
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
200

Figure F.40: BSL Projections for Shareholders
Equity
Mean = 160% 210%; Median = 93%

Figure F.41: MDM Projections for Related Party
Translations
Mean = 46% 50%

Figure F.42: MDM Projections for Auditors
Opinion
Mean = 1.2% 1.1%; Median = 0%

Figure F.43: MDM Projections for Change in
Auditors
Mean = 5% 21%; Median = 0%

Figure F.44: MDM Projections for Management
Turnover
Mean = 22% 39%; Median = 0%

Figure F.45: MDM Projections for Legal
Proceedings
Mean = 32% 46%; Median = 0%
0
50
100
150
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 > 5
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
200
250
300
350
0% 50% 100%
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
600
0% 50% 100%
F
r
e
q
u
e
n
c
y
0
100
200
300
400
500
600
0% 50% 100%
F
r
e
q
u
e
n
c
y
0
50
100
150
200
250
300
350
400
450
0% 50% 100%
F
r
e
q
u
e
n
c
y
0
100
200
300
400
0% 50% 100%
F
r
e
q
u
e
n
c
y
201

Figure F.46: MDM Projections for Retirement Plans
Mean = 14% 39%; Median = 0%
Figure F.47: Second Stage Projections for IS Metric
Mean = 120% 170%; Median = 50%

Figure F.48: Second Stage Projections for BSA
Metric
Mean = 94% 190%; Median = 33%

Figure F.49: Second Stage Projections for BSL
Metric
Mean = 95% 190%; Median = 42%

Figure F.50: Second Stage Projections for MDM
Metric
Mean = 160% 320%; Median = 20%

0
100
200
300
400
500
0% 50% 100%
F
r
e
q
u
e
n
c
y
0
40
80
120
160
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
0
50
100
150
200
250
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 > 1
F
r
e
q
u
e
n
c
y
(By Factor)
202

F.8 Second Stage Model Results with Different Time Windows

Figure F.51: Prediction by Second Stage Model from One Year Back, 1996-2008



0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Layer
Type I Error
Type II Error
Altman - Type I Error
Altman - Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
Layer
Type I Error
Type II Error
203



0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19
Layer
Type I Error
Type II Error
0%
20%
40%
60%
80%
100%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Layer
Type I Error
Type II Error
204

F.9 Probability Functions for Different Time Windows

Figure F.56: Probability of Bankruptcy from One Year Back, 1996-2009


y = 0.54x
2
- 0.81x + 0.32
R = 0.93
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data
y = 0.63x
2
- 0.91x + 0.34
R = 0.74
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data
205




y = 0.24x
2
- 0.41x + 0.20
R = 0.89
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data
y = 0.16x
2
- 0.31x + 0.17
R = 0.91
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data
y = 0.05x
2
- 0.12x + 0.10
R = 0.83
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data
206

y = -0.24x
2
+ 0.24x
R = 0.53
0%
10%
20%
30%
40%
50%
0.00 0.20 0.40 0.60 0.80 1.00
Probability
of
Bankruptcy
Actual Data

Kingyens Angela T 201211 PHD Thesis

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Kingyens Angela T 201211 PHD Thesis

Uploaded by

Copyright:

Available Formats

BANKRUPTCY PREDICTION OF

COMPANIES IN THE RETAIL-APPAREL INDUSTRY

Average Total Assets

is the expected return of the firm, R

is often referred to as the market premium.

on the frontier where:

are the coordinates of a virtual linear composite DMU (i.e. DMU

on the frontier where:

are the coordinates of a virtual linear composite DMU (i.e. DMU

has the standard normal distribution.

s ), CCR projection for G is achieved by a 33.33% reduction in input 1 and a

You might also like